Elasticity [4 ed.] 3031152131, 9783031152139

Table of contents :
Preface
Contents
Part I General Considerations
1 Introduction
1.1 Notation for Stress and Displacement
1.1.1 Stress
1.1.2 Index and vector notation and the summation convention
1.1.3 Vector operators in index notation
1.1.4 Vectors, tensors and transformation rules
1.1.5 Principal stresses and von Mises stress
1.1.6 Displacement
1.2 Strains and their Relation to Displacements
1.2.1 Tensile strain
1.2.2 Rotation and shear strain
1.2.3 Transformation of coördinates
1.2.4 Definition of shear strain
1.3 Stress-strain Relations
1.3.1 Isotropic constitutive law
1.3.2 Lamé's constants
1.3.3 Dilatation and bulk modulus
1.3.4 Deviatoric stress
Problems
2 Equilibrium and Compatibility
2.1 Equilibrium Equations
2.2 Compatibility Equations
2.2.1 The significance of the compatibility equations
2.3 Equilibrium Equations in terms of Displacements
Problems
Part II Two-dimensional Problems
3 Plane Strain and Plane Stress
3.1 Plane Strain
3.1.1 The corrective solution
3.1.2 Saint-Venant's principle
3.2 Plane Stress
3.2.1 Generalized plane stress
3.2.2 Relationship between plane stress and plane strain
Problems
4 Stress Function Formulation
4.1 Choice of a Suitable Form
4.2 The Airy Stress Function
4.2.1 Transformation of coördinates
4.2.2 Non-zero body forces
4.3 The Governing Equation
4.3.1 The compatibility condition
4.3.2 Method of solution
4.3.3 Reduced dependence on elastic constants
Problems
5 Problems in Rectangular Coördinates
5.1 Biharmonic Polynomial Functions
5.1.1 Second and third degree polynomials
5.2 Rectangular Beam Problems
5.2.1 Bending of a beam by an end load
5.2.2 Higher order polynomials — a general strategy
5.2.3 Manual solutions — symmetry considerations
5.3 Fourier Series and Transform Solutions
5.3.1 Choice of form
5.3.2 Fourier transforms
Problems
6 End Effects
6.1 Decaying Solutions
6.2 The Corrective Solution
6.2.1 Separated-variable solutions
6.2.2 The eigenvalue problem
6.3 Other Saint-Venant Problems
6.4 Mathieu's Solution
Problems
7 Body Forces
7.1 Stress Function Formulation
7.1.1 Conservative vector fields
7.1.2 The compatibility condition
7.2 Particular Cases
7.2.1 Gravitational loading
7.2.2 Inertia forces
7.2.3 Quasi-static problems
7.2.4 Rigid-body kinematics
7.3 Solution for the Stress Function
7.3.1 The rotating rectangular bar
7.3.2 Solution of the governing equation
7.4 Rotational Acceleration
7.4.1 The circular disk
7.4.2 The rectangular bar
7.4.3 Weak boundary conditions and the equation of motion
Problems
8 Problems in Polar Coördinates
8.1 Expressions for Stress Components
8.2 Strain Components
8.3 Fourier Series Expansion
8.3.1 Satisfaction of boundary conditions
8.3.2 Circular hole in a shear field
8.3.3 Degenerate cases
8.4 The Michell Solution
8.4.1 Hole in a tensile field
Problems
9 Calculation of Displacements
9.1 The Cantilever with an End Load
9.1.1 Rigid-body displacements and end conditions
9.1.2 Deflection of the free end
9.2 The Circular Hole
9.3 Displacements for the Michell Solution
9.3.1 Equilibrium considerations
9.3.2 The cylindrical pressure vessel
Problems
10 Curved Beam Problems
10.1 Loading at the Ends
10.1.1 Pure bending
10.1.2 Force transmission
10.2 Eigenvalues and Eigenfunctions
10.3 The Inhomogeneous Problem
10.3.1 Beam with sinusoidal loading
10.3.2 The near-singular problem
10.4 Some General Considerations
10.4.1 Conclusions
Problems
11 Wedge Problems
11.1 Power-law Tractions
11.1.1 Uniform tractions
11.1.2 The rectangular body revisited
11.1.3 More general uniform loading
11.1.4 Eigenvalues for the wedge angle
11.2 Williams' Asymptotic Method
11.2.1 Acceptable singularities
11.2.2 Eigenfunction expansion
11.2.3 Nature of the eigenvalues
11.2.4 The singular stress fields
11.2.5 Other geometries
11.3 General Loading of the Faces
Problems
12 Plane Contact Problems
12.1 Self-Similarity
12.2 The Flamant Solution
12.3 The Half-Plane
12.3.1 The normal force Fy
12.3.2 The tangential force Fx
12.3.3 Summary
12.4 Distributed Normal Tractions
12.5 Frictionless Contact Problems
12.5.1 Method of solution
12.5.2 The flat punch
12.5.3 The cylindrical punch (Hertz problem)
12.6 Problems with Two Deformable Bodies
12.7 Uncoupled Problems
12.7.1 Contact of cylinders
12.8 Combined Normal and Tangential Loading
12.8.1 Cattaneo and Mindlin's problem
12.8.2 Steady rolling: Carter's solution
Problems
13 Forces, Dislocations and Cracks
13.1 The Kelvin Solution
13.1.1 Body force problems
13.2 Dislocations
13.2.1 Dislocations in Materials Science
13.2.2 Similarities and differences
13.2.3 Dislocations as Green's functions
13.2.4 Stress concentrations
13.3 Crack Problems
13.3.1 Linear Elastic Fracture Mechanics
13.3.2 Plane crack in a tensile field
13.3.3 Energy release rate
13.4 Disclinations
13.4.1 Disclinations in a crystal structure
13.5 Method of Images
Problems
14 Thermoelasticity
14.1 The Governing Equation
14.2 Heat Conduction
14.3 Steady-state Problems
14.3.1 Dundurs' Theorem
Problems
15 Antiplane Shear
15.1 Transformation of Coördinates
15.2 Boundary Conditions
15.3 The Rectangular Bar
15.4 The Concentrated Line Force
15.5 The Screw Dislocation
Problems
16 Moderately Thick Plates
16.1 Boundary Conditions
16.2 Edge Effects
16.3 Body Force Problems
16.4 Normal Loading of the Faces
16.4.1 Steady-state thermoelasticity
Problems
Part III End Loading of the Prismatic Bar
17 Torsion of a Prismatic Bar
17.1 Prandtl's Stress Function
17.1.1 Solution of the governing equation
17.1.2 The warping function
17.2 The Membrane Analogy
17.3 Thin-walled Open Sections
17.4 The Rectangular Bar
17.5 Multiply-connected (Closed) Sections
17.5.1 Thin-walled closed sections
Problems
18 Shear of a Prismatic Bar
18.1 The Semi-inverse Method
18.2 Stress Function Formulation
18.3 The Boundary Condition
18.3.1 Integrability
18.3.2 Relation to the torsion problem
18.4 Methods of Solution
18.4.1 The circular bar
18.4.2 The rectangular bar
Problems
Part IV Complex-Variable Formulation
19 Prelinary Mathematical Results
19.1 Holomorphic Functions
19.2 Harmonic Functions
19.3 Biharmonic Functions
19.4 Expressing Real Harmonic and Biharmonic Functions in Complex Form
19.4.1 Biharmonic functions
19.5 Line Integrals
19.5.1 The residue theorem
19.5.2 The Cauchy integral theorem
19.6 Solution of Harmonic Boundary-value Problems
19.6.1 Direct method for the interior problem for a circle
19.6.2 Direct method for the exterior problem for a circle
19.6.3 The half-plane
19.7 Conformal Mapping
Problems
20 Application to Elasticity Problems
20.1 Representation of Vectors
20.1.1 Transformation of coördinates
20.2 The Antiplane Problem
20.2.1 Solution of antiplane boundary-value problems
20.3 In-plane Deformations
20.3.1 Expressions for stresses
20.3.2 Rigid-body displacement
20.4 Relation between the Airy Stress Function and the Complex Potentials
20.5 Boundary Tractions
20.5.1 Equilibrium considerations
20.6 Boundary-value Problems
20.6.1 Solution of the interior problem for the circle
20.6.2 Solution of the exterior problem for the circle
20.7 Conformal Mapping for In-plane Problems
20.7.1 The elliptical hole
Problems
Part V Three-Dimensional Problems
21 Displacement Function Solutions
21.1 The Strain Potential
21.2 The Galerkin Vector
21.3 The Papkovich-Neuber Solution
21.3.1 Change of coördinate system
21.4 Completeness and Uniqueness
21.4.1 Methods of partial integration
21.5 Body Forces
21.5.1 Conservative body force fields
21.5.2 Non-conservative body force fields
Problems
22 The Boussinesq Potentials
22.1 Solution A: The Strain Potential
22.2 Solution B
22.3 Solution E: Rotational Deformation
22.4 Other Coördinate Systems
22.4.1 Cylindrical polar coördinates
22.4.2 Spherical polar coördinates
22.5 Solutions Obtained by Superposition
22.5.1 Solution F: Frictionless isothermal contact problems
22.5.2 Solution G: The surface free of normal traction
22.5.3 A plane strain solution
22.6 A Three-dimensional Complex-Variable Solution
Problems
23 Thermoelastic Displacement Potentials
23.1 The Method of Strain Suppression
23.2 Boundary-value Problems
23.2.1 Spherically-symmetric Stresses
23.2.2 More general geometries
23.3 Plane Problems
23.3.1 Axisymmetric problems for the cylinder
23.3.2 Steady-state plane problems
23.3.3 Heat flow perturbed by a circular hole
23.3.4 Plane stress
23.4 Steady-state Temperature: Solution T
23.4.1 Thermoelastic plane stress
Problems
24 Singular Solutions
24.1 The Source Solution
24.1.1 The centre of dilatation
24.1.2 The Kelvin solution
24.2 Dimensional Considerations
24.2.1 The Boussinesq solution
24.3 Other Singular Solutions
24.4 Image Methods
24.4.1 The traction-free half-space
Problems
25 Spherical Harmonics
25.1 Fourier Series Solution
25.2 Reduction to Legendre's Equation
25.3 Axisymmetric Potentials and Legendre Polynomials
25.3.1 Singular spherical harmonics
25.3.2 Special cases
25.4 Non-axisymmetric Harmonics
25.5 Cartesian and Cylindrical Polar Coördinates
25.6 Harmonic Potentials with Logarithmic Terms
25.6.1 Logarithmic functions for cylinder problems
25.7 Non-axisymmetric Cylindrical Potentials
25.8 Spherical Harmonics in Complex-variable Notation
25.8.1 Bounded cylindrical harmonics
25.8.2 Singular cylindrical harmonics
Problems
26 Cylinders and Circular Plates
26.1 Axisymmetric Problems for Cylinders
26.1.1 The solid cylinder
26.1.2 The hollow cylinder
26.2 Axisymmetric Circular Plates
26.2.1 Uniformly loaded plate on a simple support
26.3 Non-axisymmetric Problems
26.3.1 Cylindrical cantilever with an end load
Problems
27 Problems in Spherical Coördinates
27.1 Solid and Hollow Spheres
27.1.1 The solid sphere in torsion
27.1.2 Spherical hole in a tensile field
27.2 Conical Bars
27.2.1 Conical bar transmitting an axial force
27.2.2 Inhomogeneous problems
27.2.3 Non-axisymmetric problems
Problems
28 Eigenstrains and Inclusions
28.1 Governing Equations
28.2 Galerkin Vector Formulation
28.2.1 Non-differentiable eigenstrains
28.2.2 The stress field
28.3 Uniform Eigenstrains in a Spherical Inclusion
28.3.1 Stresses outside the inclusion
28.4 Green's Function Solutions
28.4.1 Nuclei of strain
28.5 The Ellipsoidal Inclusion
28.5.1 The stress field
28.5.2 Anisotropic material
28.6 The Ellipsoidal Inhomogeneity
28.6.1 Equal Poisson's ratios
28.6.2 The ellipsoidal hole
28.7 Energetic Considerations
28.7.1 Evaluating the integral
28.7.2 Strain energy in the inclusion
Problems
29 Axisymmetric Torsion
29.1 The Transmitted Torque
29.2 The Governing Equation
29.3 Solution of the Governing Equation
29.4 The Displacement Field
29.5 Cylindrical and Conical Bars
29.5.1 The centre of rotation
29.6 The Saint Venant Problem
Problems
30 The Prismatic Bar
30.1 Power-series Solutions
30.1.1 Superposition by differentiation
30.1.2 The problems mathcalP0 and mathcalP1
30.1.3 Properties of the solution to mathcalPm
30.2 Solution of mathcalPm by Integration
30.3 The Integration Process
30.4 The Two-dimensional Problem mathcalP0
30.5 Problem mathcalP1
30.5.1 The corrective antiplane solution
30.5.2 The circular bar
30.6 The Corrective In-plane Solution
30.7 Corrective Solutions using Real Stress Functions
30.7.1 Airy function
30.7.2 Prandtl function
30.8 Solution Procedure
30.9 Example
30.9.1 Problem mathcalP1
30.9.2 Problem mathcalP2
30.9.3 End conditions
Problems
31 Frictionless Contact
31.1 Boundary Conditions
31.1.1 Mixed boundary-value problems
31.2 Determining the Contact Area
31.3 Contact Problems Involving Adhesive Forces
Problems
32 The Boundary-value Problem
32.1 Hankel Transform Methods
32.2 Collins' Method
32.2.1 Indentation by a flat punch
32.2.2 Integral representation
32.2.3 Basic forms and surface values
32.2.4 Reduction to an Abel equation
32.2.5 Smooth contact problems
32.2.6 Choice of form
32.3 Non-axisymmetric Problems
32.3.1 The full stress field
Problems
33 The Penny-shaped Crack
33.1 The Penny-shaped Crack in Tension
33.2 Thermoelastic Problems
Problems
34 Hertzian Contact
34.1 Elastic Deformation
34.1.1 Field-point integration
34.2 Solution Procedure
34.2.1 Axisymmetric bodies
Problems
35 The Interface Crack
35.1 The Uncracked Interface
35.2 The Corrective Solution
35.2.1 Global conditions
35.2.2 Mixed conditions
35.3 The Penny-shaped Crack in Tension
35.3.1 Reduction to a single equation
35.3.2 Oscillatory singularities
35.4 The Contact Solution
35.5 Implications for Fracture Mechanics
Problems
36 Anisotropic Elasticity
36.1 The Constitutive Law
36.2 Two-dimensional Solutions
36.3 Orthotropic Material
36.3.1 Normal loading of the half-plane
36.3.2 Degenerate cases
36.4 Lekhnitskii's Formalism
36.4.1 Polynomial solutions
36.4.2 Solutions in linearly transformed space
36.5 Stroh's Formalism
36.5.1 The eigenvalue problem
36.5.2 Solution of boundary-value problems
36.5.3 The line force solution
36.5.4 Internal forces and dislocations
36.5.5 Planar crack problems
36.5.6 The Barnett-Lothe tensors
36.6 End Loading of the Prismatic Bar
36.6.1 Bending and axial force
36.6.2 Torsion
36.7 Three-dimensional Problems
36.7.1 Concentrated force on a half-space
36.8 Transverse Isotropy
Problems
37 Variational Methods
37.1 Strain Energy
37.1.1 Strain energy density
37.2 Conservation of Energy
37.3 Potential Energy of the External Forces
37.4 Theorem of Minimum Total Potential Energy
37.5 Approximate Solutions — the Rayleigh-Ritz Method
37.6 Castigliano's Second Theorem
37.7 Approximations using Castigliano's Second Theorem
37.7.1 The torsion problem
37.7.2 The in-plane problem
37.8 Uniqueness and Existence of Solution
37.8.1 Singularities
Problems
38 The Reciprocal Theorem
38.1 Maxwell's Theorem
38.1.1 Example: Mindlin's problem
38.2 Betti's Theorem
38.2.1 Change of volume
38.2.2 A tilted punch problem
38.2.3 Indentation of a half-space
38.3 Eigenstrain Problems
38.3.1 Deformation of a traction-free body
38.3.2 Displacement constraints
38.4 Thermoelastic Problems
Problems
Appendix A Using Maple and Mathematica
Index

Citation preview

Solid Mechanics and Its Applications

J. R. Barber

Elasticity Fourth Edition

Solid Mechanics and Its Applications Founding Editor G. M. L. Gladwell

Volume 172

Series Editors J. R. Barber, Department of Mechanical Engineering, University of Michigan, Ann Arbor, MI, USA Anders Klarbring, Mechanical Engineering, Linköping University, Linköping, Sweden

The fundamental questions arising in mechanics are: Why?, How?, and How much? The aim of this series is to provide lucid accounts written by authoritative researchers giving vision and insight in answering these questions on the subject of mechanics as it relates to solids. The scope of the series covers the entire spectrum of solid mechanics. Thus it includes the foundation of mechanics; variational formulations; computational mechanics; statics, kinematics and dynamics of rigid and elastic bodies; vibrations of solids and structures; dynamical systems and chaos; the theories of elasticity, plasticity and viscoelasticity; composite materials; rods, beams, shells and membranes; structural control and stability; soils, rocks and geomechanics; fracture; tribology; experimental mechanics; biomechanics and machine design. The median level of presentation is the first year graduate student. Some texts are monographs defining the current state of the field; others are accessible to final year undergraduates; but essentially the emphasis is on readability and clarity. Springer and Professors Barber and Klarbring welcome book ideas from authors. Potential authors who wish to submit a book proposal should contact Dr. Mayra Castro, Senior Editor, Springer Heidelberg, Germany, email: [email protected] Indexed by SCOPUS, Ei Compendex, EBSCO Discovery Service, OCLC, ProQuest Summon, Google Scholar and SpringerLink.

J. R. Barber

Elasticity Fourth Edition

J. R. Barber Department of Mechanical Engineering University of Michigan Ann Arbor, MI, USA

ISSN 0925-0042 ISSN 2214-7764 (electronic) Solid Mechanics and Its Applications ISBN 978-3-031-15213-9 ISBN 978-3-031-15214-6 (eBook) https://doi.org/10.1007/978-3-031-15214-6 1st edition: © Springer Netherlands, Kluwer Academics Publishers 1992 2nd & 3rd editions: © Springer Netherlands, Springer Science+Business Media B.V. 2004, 2010 4th edition: © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors, and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Switzerland AG The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

Preface

The subject of Elasticity can be approached from several points of view, depending on whether the practitioner is principally interested in the mathematical structure of the subject or in its use in engineering applications and, in the latter case, whether essentially numerical or analytical methods are envisaged as the solution method. My first introduction to the subject was in response to a need for information about a specific problem in Tribology. As a practising Engineer with a background only in elementary Mechanics of Materials, I approached that problem initially using the concepts of concentrated forces and superposition. Today, with a rather more extensive knowledge of analytical techniques in Elasticity, I still find it helpful to go back to these roots in the elementary theory and think through a problem physically as well as mathematically, whenever some new and unexpected feature presents difficulties in research. This way of thinking will be found to permeate this book. My engineering background will also reveal itself in a tendency to work examples through to final expressions for stresses and displacements, rather than leave the derivation at a point where the remaining manipulations would be mathematically routine. The first edition of this book, published in 1992, was based on a one semester graduate course on Linear Elasticity that I have taught at the University of Michigan since 1983. In three subsequent revisions, the amount of material has more than doubled and the character of the book has necessarily changed, but I remain committed to my original objective of writing for those who wish to find the solution of specific practical engineering problems. With this in mind, I have endeavoured to keep to a minimum any dependence on previous knowledge of Solid Mechanics, Continuum Mechanics or Mathematics. Most of the text should be readily intelligible to a reader with an undergraduate background of one or two courses in elementary Mechanics of Materials and a rudimentary knowledge of partial differentiation. Cartesian tensor notation and the summation convention are used in a few places to shorten the derivation of some general results, but these sections are carefully explained, so as to be self-explanatory.

v

vi

Preface

Modern practitioners of Elasticity are necessarily influenced by developments in numerical methods, which promise to solve all problems with no more information about the subject than is needed to formulate the description of a representative element of material in a relatively simple state of stress. As a researcher in Solid Mechanics, with a primary interest in the physical behaviour of the systems I am investigating, rather than in the mathematical structure of the solutions, I have frequently had recourse to numerical methods of all types and have tended to adopt the pragmatic criterion that the best method is that which gives the most convincing and accurate result in the shortest time. In this context, ‘convincing’ means that the solution should be capable of being checked against reliable closed-form solutions in suitable limiting cases and that it is demonstrably stable and in some sense convergent. Measured against these criteria, the ‘best’ solution to many practical problems is often not a direct numerical method, such as the finite element method, but rather one involving some significant analytical steps before the final numerical evaluation. This is particularly true in three-dimensional problems, where direct numerical methods are extremely computer-intensive if any reasonably accuracy is required, and in problems involving infinite or semi-infinite domains, discontinuities, bonded or contacting material interfaces or theoretically singular stress fields. By contrast, I would immediately opt for a finite element solution of any two-dimensional problem involving finite bodies with relatively smooth contours, unless it happened to fall into the (surprisingly wide) class of problems to which the solution can be written down in closed form. The reader will therefore find my choice of topics significantly biassed towards those fields identified above where analytical methods are most useful. As in the earlier editions, I encourage the reader to become familiar with the use of symbolic mathematical languages such as MapleTM and MathematicaTM , since these tools open up the possibility of solving considerably more complex and hence interesting and realistic elasticity problems. They also enable the student to focus on the formulation of the problem (e.g. the appropriate governing equations and boundary conditions) rather than on the algebraic manipulations, with a consequent improvement in insight into the subject and in motivation. Finally, they each posess post-processing graphics facilities that enable the user to explore important features of the resulting stress state. The reader can access numerous files in the https:// link.springer.com/book/10.1007/978-90-481-3809-8 Supplementary Material or at my University of Michigan homepage http://www-personal.umich.edu/jbarber/ela sticity/book4.html, including the solution of sample problems, electronic versions of the tables in Chapters 22 and 23 algorithms for the generation of spherical harmonic potentials. Some hints about the use of this material are contained in Appendix A, and more detailed tips about programming are included at the above websites. Those who have never used these methods will find that it takes only a few hours of trial and error to learn how to write programs to solve boundary-value problems in elasticity. This new edition contains four additional chapters, including one on the Eshelby inclusion problem, and a brief introduction to anisotropic elasticity. Also discussed are Love’s theory of ‘moderately thick plates’ and the three-dimensional Hertz contact problem.

Preface

vii

The new edition contains numerous additional end-of-chapter problems. As with previous editions, a full set of solutions to these problems is available to bona fide instructors on request to the author. Some of these problems are quite challenging, indeed several were the subject of substantial technical papers within the not too distant past, but they can all be solved in a few hours using Maple or Mathematica. Many texts on Elasticity contain problems which offer a candidate stress function and invite the student to ‘verify’ that it defines the solution to a given problem. Students invariably raise the question ‘How would we know to choose that form if we were not given it in advance?’ I have tried wherever possible to avoid this by expressing the problems in the form they would arise in Engineering—i.e. as a body of a given geometry subjected to prescribed loading. This in turn has required me to write the text in such a way that the student can approach problems deductively. I have also generally opted for explaining difficulties that might arise in an ‘obvious’ approach to the problem, rather than steering the reader around them in the interests of brevity or elegance of solution. I have taken this opportunity to correct the numerous typographical errors in the third edition, but no doubt despite my best efforts, the new material will contain more. Please communicate any errors to me. As in previous editions, I would like to thank my graduate students and more generally scientific correspondents worldwide whose questions continue to force me to re-examine my knowledge of the subject. I am also grateful to Professor John Dundurs for permission to use Table 9.1 and to the Royal Society of London for permission to reproduce Figures 13.2, 13.3. Ann Arbor, USA 2022

J. R. Barber

Contents

Part I 1

2

General Considerations

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1 Notation for Stress and Displacement . . . . . . . . . . . . . . . . . . . . . . . 1.1.1 Stress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1.2 Index and vector notation and the summation convention . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1.3 Vector operators in index notation . . . . . . . . . . . . . . . . . . . 1.1.4 Vectors, tensors and transformation rules . . . . . . . . . . . . . 1.1.5 Principal stresses and von Mises stress . . . . . . . . . . . . . . . 1.1.6 Displacement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Strains and their Relation to Displacements . . . . . . . . . . . . . . . . . . 1.2.1 Tensile strain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.2 Rotation and shear strain . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.3 Transformation of coördinates . . . . . . . . . . . . . . . . . . . . . . 1.2.4 Definition of shear strain . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Stress-strain Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3.1 Isotropic constitutive law . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3.2 Lamé’s constants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3.3 Dilatation and bulk modulus . . . . . . . . . . . . . . . . . . . . . . . . 1.3.4 Deviatoric stress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3 4 4 6 8 9 13 14 15 15 17 18 20 21 22 23 24 25 25

Equilibrium and Compatibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Equilibrium Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Compatibility Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.1 The significance of the compatibility equations . . . . . . . . 2.3 Equilibrium Equations in terms of Displacements . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

27 27 28 30 34 35

ix

x

Contents

Part II

Two-dimensional Problems

3

Plane Strain and Plane Stress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Plane Strain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1.1 The corrective solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1.2 Saint-Venant’s principle . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Plane Stress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.1 Generalized plane stress . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.2 Relationship between plane stress and plane strain . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

41 41 43 44 45 46 46 47

4

Stress Function Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1 Choice of a Suitable Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 The Airy Stress Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.1 Transformation of coördinates . . . . . . . . . . . . . . . . . . . . . . 4.2.2 Non-zero body forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3 The Governing Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3.1 The compatibility condition . . . . . . . . . . . . . . . . . . . . . . . . 4.3.2 Method of solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3.3 Reduced dependence on elastic constants . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

51 51 52 53 54 54 55 55 56 56

5

Problems in Rectangular Coördinates . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1 Biharmonic Polynomial Functions . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1.1 Second and third degree polynomials . . . . . . . . . . . . . . . . 5.2 Rectangular Beam Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.1 Bending of a beam by an end load . . . . . . . . . . . . . . . . . . . 5.2.2 Higher order polynomials — a general strategy . . . . . . . 5.2.3 Manual solutions — symmetry considerations . . . . . . . . 5.3 Fourier Series and Transform Solutions . . . . . . . . . . . . . . . . . . . . . . 5.3.1 Choice of form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.2 Fourier transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

61 61 63 65 65 67 72 75 75 79 80

6

End Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.1 Decaying Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2 The Corrective Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2.1 Separated-variable solutions . . . . . . . . . . . . . . . . . . . . . . . . 6.2.2 The eigenvalue problem . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3 Other Saint-Venant Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.4 Mathieu’s Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

83 83 84 85 86 90 91 95

Contents

7

xi

Body Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.1 Stress Function Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.1.1 Conservative vector fields . . . . . . . . . . . . . . . . . . . . . . . . . . 7.1.2 The compatibility condition . . . . . . . . . . . . . . . . . . . . . . . . 7.2 Particular Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2.1 Gravitational loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2.2 Inertia forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2.3 Quasi-static problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2.4 Rigid-body kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3 Solution for the Stress Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3.1 The rotating rectangular bar . . . . . . . . . . . . . . . . . . . . . . . . 7.3.2 Solution of the governing equation . . . . . . . . . . . . . . . . . . 7.4 Rotational Acceleration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.4.1 The circular disk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.4.2 The rectangular bar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.4.3 Weak boundary conditions and the equation of motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

97 97 98 99 99 100 100 101 101 103 103 105 106 107 109

8

Problems in Polar Coördinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.1 Expressions for Stress Components . . . . . . . . . . . . . . . . . . . . . . . . . 8.2 Strain Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.3 Fourier Series Expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.3.1 Satisfaction of boundary conditions . . . . . . . . . . . . . . . . . 8.3.2 Circular hole in a shear field . . . . . . . . . . . . . . . . . . . . . . . . 8.3.3 Degenerate cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.4 The Michell Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.4.1 Hole in a tensile field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

117 117 119 120 121 121 124 126 128 129

9

Calculation of Displacements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.1 The Cantilever with an End Load . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.1.1 Rigid-body displacements and end conditions . . . . . . . . . 9.1.2 Deflection of the free end . . . . . . . . . . . . . . . . . . . . . . . . . . 9.2 The Circular Hole . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.3 Displacements for the Michell Solution . . . . . . . . . . . . . . . . . . . . . . 9.3.1 Equilibrium considerations . . . . . . . . . . . . . . . . . . . . . . . . . 9.3.2 The cylindrical pressure vessel . . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

133 133 135 137 137 140 141 142 143

10 Curved Beam Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10.1 Loading at the Ends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10.1.1 Pure bending . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10.1.2 Force transmission . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10.2 Eigenvalues and Eigenfunctions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10.3 The Inhomogeneous Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

147 147 147 150 152 152

112 112

xii

Contents

10.3.1 Beam with sinusoidal loading . . . . . . . . . . . . . . . . . . . . . . 10.3.2 The near-singular problem . . . . . . . . . . . . . . . . . . . . . . . . . 10.4 Some General Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10.4.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

153 156 159 159 160

11 Wedge Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 Power-law Tractions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1.1 Uniform tractions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1.2 The rectangular body revisited . . . . . . . . . . . . . . . . . . . . . . 11.1.3 More general uniform loading . . . . . . . . . . . . . . . . . . . . . . 11.1.4 Eigenvalues for the wedge angle . . . . . . . . . . . . . . . . . . . . 11.2 Williams’ Asymptotic Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2.1 Acceptable singularities . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2.2 Eigenfunction expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2.3 Nature of the eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2.4 The singular stress fields . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2.5 Other geometries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.3 General Loading of the Faces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

163 163 164 166 167 168 169 169 171 173 176 178 180 181

12 Plane Contact Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.1 Self-Similarity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.2 The Flamant Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.3 The Half-Plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.3.1 The normal force Fy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.3.2 The tangential force Fx . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.3.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.4 Distributed Normal Tractions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.5 Frictionless Contact Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.5.1 Method of solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.5.2 The flat punch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.5.3 The cylindrical punch (Hertz problem) . . . . . . . . . . . . . . . 12.6 Problems with Two Deformable Bodies . . . . . . . . . . . . . . . . . . . . . 12.7 Uncoupled Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.7.1 Contact of cylinders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.8 Combined Normal and Tangential Loading . . . . . . . . . . . . . . . . . . . 12.8.1 Cattaneo and Mindlin’s problem . . . . . . . . . . . . . . . . . . . . 12.8.2 Steady rolling: Carter’s solution . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

185 185 186 187 188 189 189 190 191 192 194 195 198 201 201 202 203 206 209

13 Forces, Dislocations and Cracks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13.1 The Kelvin Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13.1.1 Body force problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13.2 Dislocations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13.2.1 Dislocations in Materials Science . . . . . . . . . . . . . . . . . . .

215 215 217 218 220

Contents

xiii

13.2.2 Similarities and differences . . . . . . . . . . . . . . . . . . . . . . . . . 13.2.3 Dislocations as Green’s functions . . . . . . . . . . . . . . . . . . . 13.2.4 Stress concentrations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13.3 Crack Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13.3.1 Linear Elastic Fracture Mechanics . . . . . . . . . . . . . . . . . . . 13.3.2 Plane crack in a tensile field . . . . . . . . . . . . . . . . . . . . . . . . 13.3.3 Energy release rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13.4 Disclinations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13.4.1 Disclinations in a crystal structure . . . . . . . . . . . . . . . . . . . 13.5 Method of Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

221 222 222 223 224 225 228 230 231 232 235

14 Thermoelasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14.1 The Governing Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14.2 Heat Conduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14.3 Steady-state Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14.3.1 Dundurs’ Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

239 239 241 242 243 245

15 Antiplane Shear . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15.1 Transformation of Coördinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15.2 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15.3 The Rectangular Bar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15.4 The Concentrated Line Force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15.5 The Screw Dislocation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

247 248 248 249 250 252 252

16 Moderately Thick Plates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16.1 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16.2 Edge Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16.3 Body Force Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16.4 Normal Loading of the Faces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16.4.1 Steady-state thermoelasticity . . . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

257 258 260 261 263 264 265

Part III End Loading of the Prismatic Bar 17 Torsion of a Prismatic Bar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17.1 Prandtl’s Stress Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17.1.1 Solution of the governing equation . . . . . . . . . . . . . . . . . . 17.1.2 The warping function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17.2 The Membrane Analogy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17.3 Thin-walled Open Sections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17.4 The Rectangular Bar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17.5 Multiply-connected (Closed) Sections . . . . . . . . . . . . . . . . . . . . . . . 17.5.1 Thin-walled closed sections . . . . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

269 270 271 273 273 274 276 277 279 281

xiv

Contents

18 Shear of a Prismatic Bar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18.1 The Semi-inverse Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18.2 Stress Function Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18.3 The Boundary Condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18.3.1 Integrability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18.3.2 Relation to the torsion problem . . . . . . . . . . . . . . . . . . . . . 18.4 Methods of Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18.4.1 The circular bar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18.4.2 The rectangular bar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

289 289 290 291 291 292 293 293 294 297

Part IV Complex-Variable Formulation 19 Prelinary Mathematical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19.1 Holomorphic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19.2 Harmonic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19.3 Biharmonic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19.4 Expressing Real Harmonic and Biharmonic Functions in Complex Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19.4.1 Biharmonic functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19.5 Line Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19.5.1 The residue theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19.5.2 The Cauchy integral theorem . . . . . . . . . . . . . . . . . . . . . . . 19.6 Solution of Harmonic Boundary-value Problems . . . . . . . . . . . . . . 19.6.1 Direct method for the interior problem for a circle . . . . . 19.6.2 Direct method for the exterior problem for a circle . . . . . 19.6.3 The half-plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19.7 Conformal Mapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 Application to Elasticity Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20.1 Representation of Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20.1.1 Transformation of coördinates . . . . . . . . . . . . . . . . . . . . . . 20.2 The Antiplane Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20.2.1 Solution of antiplane boundary-value problems . . . . . . . 20.3 In-plane Deformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20.3.1 Expressions for stresses . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20.3.2 Rigid-body displacement . . . . . . . . . . . . . . . . . . . . . . . . . . 20.4 Relation between the Airy Stress Function and the Complex Potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20.5 Boundary Tractions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20.5.1 Equilibrium considerations . . . . . . . . . . . . . . . . . . . . . . . . . 20.6 Boundary-value Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20.6.1 Solution of the interior problem for the circle . . . . . . . . . 20.6.2 Solution of the exterior problem for the circle . . . . . . . . . 20.7 Conformal Mapping for In-plane Problems . . . . . . . . . . . . . . . . . . .

301 302 303 303 305 306 306 307 310 312 312 314 315 317 319 323 323 324 325 326 328 330 331 332 334 336 337 338 340 343

Contents

xv

20.7.1 The elliptical hole . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348 Part V

Three-Dimensional Problems

21 Displacement Function Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21.1 The Strain Potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21.2 The Galerkin Vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21.3 The Papkovich-Neuber Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21.3.1 Change of coördinate system . . . . . . . . . . . . . . . . . . . . . . . 21.4 Completeness and Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21.4.1 Methods of partial integration . . . . . . . . . . . . . . . . . . . . . . 21.5 Body Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21.5.1 Conservative body force fields . . . . . . . . . . . . . . . . . . . . . . 21.5.2 Non-conservative body force fields . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

353 353 354 356 357 358 359 362 362 363 363

22 The Boussinesq Potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22.1 Solution A: The Strain Potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22.2 Solution B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22.3 Solution E: Rotational Deformation . . . . . . . . . . . . . . . . . . . . . . . . . 22.4 Other Coördinate Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22.4.1 Cylindrical polar coördinates . . . . . . . . . . . . . . . . . . . . . . . 22.4.2 Spherical polar coördinates . . . . . . . . . . . . . . . . . . . . . . . . . 22.5 Solutions Obtained by Superposition . . . . . . . . . . . . . . . . . . . . . . . . 22.5.1 Solution F: Frictionless isothermal contact problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22.5.2 Solution G: The surface free of normal traction . . . . . . . 22.5.3 A plane strain solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22.6 A Three-dimensional Complex-Variable Solution . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

365 366 366 367 368 368 369 371

23 Thermoelastic Displacement Potentials . . . . . . . . . . . . . . . . . . . . . . . . . . 23.1 The Method of Strain Suppression . . . . . . . . . . . . . . . . . . . . . . . . . . 23.2 Boundary-value Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23.2.1 Spherically-symmetric Stresses . . . . . . . . . . . . . . . . . . . . . 23.2.2 More general geometries . . . . . . . . . . . . . . . . . . . . . . . . . . . 23.3 Plane Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23.3.1 Axisymmetric problems for the cylinder . . . . . . . . . . . . . 23.3.2 Steady-state plane problems . . . . . . . . . . . . . . . . . . . . . . . . 23.3.3 Heat flow perturbed by a circular hole . . . . . . . . . . . . . . . 23.3.4 Plane stress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23.4 Steady-state Temperature: Solution T . . . . . . . . . . . . . . . . . . . . . . . 23.4.1 Thermoelastic plane stress . . . . . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

379 381 382 382 383 384 384 385 387 390 390 393 394

371 374 374 375 376

xvi

Contents

24 Singular Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24.1 The Source Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24.1.1 The centre of dilatation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24.1.2 The Kelvin solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24.2 Dimensional Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24.2.1 The Boussinesq solution . . . . . . . . . . . . . . . . . . . . . . . . . . . 24.3 Other Singular Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24.4 Image Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24.4.1 The traction-free half-space . . . . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

397 397 398 399 401 401 404 406 407 409

25 Spherical Harmonics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25.1 Fourier Series Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25.2 Reduction to Legendre’s Equation . . . . . . . . . . . . . . . . . . . . . . . . . . 25.3 Axisymmetric Potentials and Legendre Polynomials . . . . . . . . . . . 25.3.1 Singular spherical harmonics . . . . . . . . . . . . . . . . . . . . . . . 25.3.2 Special cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25.4 Non-axisymmetric Harmonics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25.5 Cartesian and Cylindrical Polar Coördinates . . . . . . . . . . . . . . . . . . 25.6 Harmonic Potentials with Logarithmic Terms . . . . . . . . . . . . . . . . . 25.6.1 Logarithmic functions for cylinder problems . . . . . . . . . . 25.7 Non-axisymmetric Cylindrical Potentials . . . . . . . . . . . . . . . . . . . . 25.8 Spherical Harmonics in Complex-variable Notation . . . . . . . . . . . 25.8.1 Bounded cylindrical harmonics . . . . . . . . . . . . . . . . . . . . . 25.8.2 Singular cylindrical harmonics . . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

411 412 412 413 414 415 415 416 417 419 420 422 423 425 425

26 Cylinders and Circular Plates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26.1 Axisymmetric Problems for Cylinders . . . . . . . . . . . . . . . . . . . . . . . 26.1.1 The solid cylinder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26.1.2 The hollow cylinder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26.2 Axisymmetric Circular Plates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26.2.1 Uniformly loaded plate on a simple support . . . . . . . . . . . 26.3 Non-axisymmetric Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26.3.1 Cylindrical cantilever with an end load . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

427 427 428 432 434 434 437 437 439

27 Problems in Spherical Coördinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27.1 Solid and Hollow Spheres . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27.1.1 The solid sphere in torsion . . . . . . . . . . . . . . . . . . . . . . . . . 27.1.2 Spherical hole in a tensile field . . . . . . . . . . . . . . . . . . . . . . 27.2 Conical Bars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27.2.1 Conical bar transmitting an axial force . . . . . . . . . . . . . . . 27.2.2 Inhomogeneous problems . . . . . . . . . . . . . . . . . . . . . . . . . . 27.2.3 Non-axisymmetric problems . . . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

443 443 444 445 447 447 450 451 453

Contents

xvii

28 Eigenstrains and Inclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28.1 Governing Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28.2 Galerkin Vector Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28.2.1 Non-differentiable eigenstrains . . . . . . . . . . . . . . . . . . . . . 28.2.2 The stress field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28.3 Uniform Eigenstrains in a Spherical Inclusion . . . . . . . . . . . . . . . . 28.3.1 Stresses outside the inclusion . . . . . . . . . . . . . . . . . . . . . . . 28.4 Green’s Function Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28.4.1 Nuclei of strain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28.5 The Ellipsoidal Inclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28.5.1 The stress field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28.5.2 Anisotropic material . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28.6 The Ellipsoidal Inhomogeneity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28.6.1 Equal Poisson’s ratios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28.6.2 The ellipsoidal hole . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28.7 Energetic Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28.7.1 Evaluating the integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28.7.2 Strain energy in the inclusion . . . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

457 458 459 460 461 462 463 463 464 465 468 468 469 470 470 471 472 472 473

29 Axisymmetric Torsion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29.1 The Transmitted Torque . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29.2 The Governing Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29.3 Solution of the Governing Equation . . . . . . . . . . . . . . . . . . . . . . . . . 29.4 The Displacement Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29.5 Cylindrical and Conical Bars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29.5.1 The centre of rotation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29.6 The Saint Venant Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

475 476 476 477 479 480 482 482 483

30 The Prismatic Bar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30.1 Power-series Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30.1.1 Superposition by differentiation . . . . . . . . . . . . . . . . . . . . . 30.1.2 The problems P0 and P1 . . . . . . . . . . . . . . . . . . . . . . . . . . . 30.1.3 Properties of the solution to Pm . . . . . . . . . . . . . . . . . . . . . 30.2 Solution of Pm by Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30.3 The Integration Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30.4 The Two-dimensional Problem P0 . . . . . . . . . . . . . . . . . . . . . . . . . . 30.5 Problem P1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30.5.1 The corrective antiplane solution . . . . . . . . . . . . . . . . . . . . 30.5.2 The circular bar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30.6 The Corrective In-plane Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . 30.7 Corrective Solutions using Real Stress Functions . . . . . . . . . . . . . . 30.7.1 Airy function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30.7.2 Prandtl function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30.8 Solution Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

485 485 486 487 488 488 489 491 492 493 494 495 496 496 497 498

xviii

Contents

30.9 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30.9.1 Problem P1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30.9.2 Problem P2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30.9.3 End conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

499 499 500 503 504

31 Frictionless Contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31.1 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31.1.1 Mixed boundary-value problems . . . . . . . . . . . . . . . . . . . . 31.2 Determining the Contact Area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31.3 Contact Problems Involving Adhesive Forces . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

507 507 508 509 512 514

32 The Boundary-value Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32.1 Hankel Transform Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32.2 Collins’ Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32.2.1 Indentation by a flat punch . . . . . . . . . . . . . . . . . . . . . . . . . 32.2.2 Integral representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32.2.3 Basic forms and surface values . . . . . . . . . . . . . . . . . . . . . 32.2.4 Reduction to an Abel equation . . . . . . . . . . . . . . . . . . . . . . 32.2.5 Smooth contact problems . . . . . . . . . . . . . . . . . . . . . . . . . . 32.2.6 Choice of form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32.3 Non-axisymmetric Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32.3.1 The full stress field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

517 517 518 518 521 522 524 527 529 530 532 534

33 The Penny-shaped Crack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33.1 The Penny-shaped Crack in Tension . . . . . . . . . . . . . . . . . . . . . . . . 33.2 Thermoelastic Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

537 537 540 543

34 Hertzian Contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34.1 Elastic Deformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34.1.1 Field-point integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34.2 Solution Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34.2.1 Axisymmetric bodies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

545 547 548 550 551 553

35 The Interface Crack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35.1 The Uncracked Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35.2 The Corrective Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35.2.1 Global conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35.2.2 Mixed conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35.3 The Penny-shaped Crack in Tension . . . . . . . . . . . . . . . . . . . . . . . . 35.3.1 Reduction to a single equation . . . . . . . . . . . . . . . . . . . . . . 35.3.2 Oscillatory singularities . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35.4 The Contact Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

555 555 557 557 558 561 563 564 564

Contents

xix

35.5 Implications for Fracture Mechanics . . . . . . . . . . . . . . . . . . . . . . . . 566 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 567 36 Anisotropic Elasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36.1 The Constitutive Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36.2 Two-dimensional Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36.3 Orthotropic Material . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36.3.1 Normal loading of the half-plane . . . . . . . . . . . . . . . . . . . . 36.3.2 Degenerate cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36.4 Lekhnitskii’s Formalism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36.4.1 Polynomial solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36.4.2 Solutions in linearly transformed space . . . . . . . . . . . . . . 36.5 Stroh’s Formalism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36.5.1 The eigenvalue problem . . . . . . . . . . . . . . . . . . . . . . . . . . . 36.5.2 Solution of boundary-value problems . . . . . . . . . . . . . . . . 36.5.3 The line force solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36.5.4 Internal forces and dislocations . . . . . . . . . . . . . . . . . . . . . 36.5.5 Planar crack problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36.5.6 The Barnett-Lothe tensors . . . . . . . . . . . . . . . . . . . . . . . . . . 36.6 End Loading of the Prismatic Bar . . . . . . . . . . . . . . . . . . . . . . . . . . . 36.6.1 Bending and axial force . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36.6.2 Torsion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36.7 Three-dimensional Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36.7.1 Concentrated force on a half-space . . . . . . . . . . . . . . . . . . 36.8 Transverse Isotropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

569 569 570 571 573 574 575 576 577 579 581 582 583 584 585 586 587 587 588 588 589 590 592

37 Variational Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37.1 Strain Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37.1.1 Strain energy density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37.2 Conservation of Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37.3 Potential Energy of the External Forces . . . . . . . . . . . . . . . . . . . . . . 37.4 Theorem of Minimum Total Potential Energy . . . . . . . . . . . . . . . . . 37.5 Approximate Solutions — the Rayleigh-Ritz Method . . . . . . . . . . 37.6 Castigliano’s Second Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37.7 Approximations using Castigliano’s Second Theorem . . . . . . . . . 37.7.1 The torsion problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37.7.2 The in-plane problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37.8 Uniqueness and Existence of Solution . . . . . . . . . . . . . . . . . . . . . . . 37.8.1 Singularities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

595 596 596 597 598 598 599 602 603 604 606 608 609 611

xx

Contents

38 The Reciprocal Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38.1 Maxwell’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38.1.1 Example: Mindlin’s problem . . . . . . . . . . . . . . . . . . . . . . . 38.2 Betti’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38.2.1 Change of volume . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38.2.2 A tilted punch problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38.2.3 Indentation of a half-space . . . . . . . . . . . . . . . . . . . . . . . . . 38.3 Eigenstrain Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38.3.1 Deformation of a traction-free body . . . . . . . . . . . . . . . . . 38.3.2 Displacement constraints . . . . . . . . . . . . . . . . . . . . . . . . . . 38.4 Thermoelastic Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

615 615 616 617 618 620 622 624 624 625 626 627

Appendix A: Using Maple and Mathematica . . . . . . . . . . . . . . . . . . . . . . . . . 631 Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 633

Part I

General Considerations

Chapter 1

Introduction

The subject of Elasticity is concerned with the determination of the stresses and displacements in a body as a result of applied mechanical or thermal loads, for those cases in which the body reverts to its original state on the removal of the loads. In this book, we shall further restrict attention to the case of linear infinitesimal elasticity, in which the stresses and displacements are linearly proportional to the applied loads, and the displacement gradients are small compared with unity. These restrictions ensure that linear superposition can be used and enable us to employ a wide range of series and transform techniques which are not available for non-linear problems. Most engineers first encounter problems of this kind in the context of the subject known as Mechanics of Materials, which is an important constituent of most undergraduate engineering curricula. Mechanics of Materials differs from Elasticity in that various plausible but unsubstantiated assumptions are made about the deformation process in the course of the analysis. A typical example is the assumption that plane sections remain plane in the bending of a slender beam. Elasticity makes no such assumptions, but attempts to develop the solution directly and rigorously from its first principles, which are Newton’s laws of motion, Euclidian geometry and Hooke’s law. Approximations are often introduced towards the end of the solution, but these are mathematical approximations used to obtain solutions of the governing equations rather than physical approximations that impose artificial and strictly unjustifiable constraints on the permissible deformation field. However, it would be a mistake to draw too firm a distinction between the two approaches, since practitioners of each have much to learn from the other. Mechanics of Materials, with its emphasis on physical reasoning and a full exploration of the practical consequences of the results, is often able to provide insights into the problem that are less easily obtained from a purely mathematical perspective. Indeed, we shall Supplementary Information The online version contains supplementary material available at https://doi.org/10.1007/978-3-031-15214-6_1. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 J. R. Barber, Elasticity, Solid Mechanics and Its Applications 172, https://doi.org/10.1007/978-3-031-15214-6_1

3

4

1 Introduction

make extensive use of physical parallels in this book and pursue many problems to conclusions relevant to practical applications, with the hope of deepening the reader’s understanding of the underlying structure of the subject. Conversely, the mathematical rigour of Elasticity gives us greater confidence in the results, since, even when we have to resort to an approximate solution, we can usually estimate its accuracy with some confidence — something that is very difficult to do with the physical approximations used in Mechanics of Materials1 . Also, there is little to be said for using an ad hoc approach when, as is often the case, a more rigorous treatment presents no serious difficulty.

1.1 Notation for Stress and Displacement It is assumed that the reader is more or less familiar with the concept of stress and strain from elementary courses on Mechanics of Materials. This section is intended to introduce the notation used, to refresh the reader’s memory about some important ideas, and to record some elementary but useful results.

1.1.1 Stress Components of stress will all be denoted by the symbol σ with appropriate suffices. The second suffix denotes the direction of the stress component and the first the direction of the outward normal to the surface upon which it acts. This notation is illustrated in Figure 1.1 for the Cartesian coördinate system x, y, z.

Fig. 1.1 Notation for stress components

Notice that one consequence of this notation is that normal (i.e. tensile and compressive) stresses have both suffices the same (e.g. σx x , σ yy , σzz in Figure 1.1) and 1

In fact, the only practical way to examine the effect of these approximations is to relax them, by considering the same problem, or maybe a simpler problem with similar features, in the context of the theory of Elasticity.

1.1 Notation for Stress and Displacement

5

are positive when tensile. The remaining six stress components in Figure 1.1 (i.e. σx y , σ yx , σ yz , σzy , σzx , σx z ) have two different suffices and are shear stresses. Books on Mechanics of Materials often use the symbol τ for shear stress, whilst retaining σ for normal stress. However, there is no need for a different symbol, since the suffices enable us to distinguish normal from shear stress components. Also, we shall find that the use of a single symbol with appropriate suffices permits matrix methods to be used in many derivations and introduces considerable economies in the notation for general results. The equilibrium of moments acting on the block in Figure 1.1 requires that (1.1) σx y = σ yx ; σ yz = σzy and σzx = σx z . This has the incidental advantage of rendering mistakes about the order of suffices harmless! (In fact, a few books use the opposite convention.) Readers who have not encountered three-dimensional problems before should note that there are two shear stress components on each surface and one normal stress component. There are some circumstances in which it is convenient to combine the two shear stresses on a given plane into a two-dimensional vector in the plane — i.e. to refer to the resultant shear stress on the plane. An elementary situation where this is helpful is in the Mechanics of Materials problem of determining the distribution of shear stress on the cross section of a beam due to a transverse shear force2 . For example, we note that in this case, the resultant shear stress on the plane must be tangential to the edge at the boundary of the cross section, since the shear stress complementary to the component normal to the edge acts on the traction-free surface of the beam and must therefore be zero. This of course is why the shear stress in a thin-walled section tends to follow the direction of the wall. We shall refer to a plane normal to the x-direction as an ‘x-plane’ etc. The only stress components which act on an x-plane are those which have an x as the first suffix (This is an immediate consequence of the definition). Notice also that any x-plane can be defined by an equation of the form x = c, where c is a constant. More precisely, we can define a ‘positive x-plane’ as a plane for which the positive x-direction is the outward normal and a ‘negative x-plane’ as one for which it is the inward normal. This distinction can be expressed mathematically in terms of inequalities. Thus, if part of the boundary of a solid is expressible as x = c, the solid must locally occupy one side or the other of this plane. If the domain of the solid is locally described by x < c, the bounding surface is a positive x-plane, whereas if it is described by x > c, the bounding surface is a negative x-plane. The discussion in this section suggests a useful formalism for correctly defining the boundary conditions in problems where the boundaries are parallel to the coördinate axes. We first identify the equations which define the boundaries of the solid and then write down the three traction components which act on each boundary. For example, suppose we have a rectangular solid defined by the inequalities 0 < x < a,

2

For the elasticity solution of this problem, see Chapter 18.

6

1 Introduction

0 < y < b, 0 < z < c. It is clear that the surface y = b is a positive y-plane and we deduce immediately that the corresponding traction boundary conditions will involve the stress components σ yx , σ yy , σ yz — i.e. the three components that have y as the first suffix. This procedure insures against the common student mistake of assuming (for example) that the component σx x must be zero if the surface y = b is to be traction free. (Note : Don’t assume that this mistake is too obvious for you to fall into it. When the problem is geometrically or algebraically very complicated, it is only too easy to get distracted.) Stress components can be defined in the same way for other systems of orthogonal coördinates. For example, components for the system of cylindrical polar coördinates (r, θ, z) are shown in Figure 1.2.

Fig. 1.2 Stress components in polar coördinates.

(This is a case where the definition of the ‘θ-plane’ through an equation, θ = α, is easier to comprehend than ‘the plane normal to the θ-direction’. However, note that the θ-direction is the direction in which a particle would move if θ were increased with r, z constant.)

1.1.2 Index and vector notation and the summation convention In expressing or deriving general results, it is often convenient to make use of vector notation. We shall use bold face symbols to represent vectors and single suffices to define their components in a given coördinate direction. Thus, the vector V can be written T  , (1.2) V = i Vx + j Vy + kVz = Vx , Vy , Vz where i, j , k are unit vectors in directions x, y, z respectively. In the last expression in (1.2), we have used the convenient linear algebra notation for a vector, but to take full advantage of this, we also need to replace Vx , Vy , Vz by V1 , V2 , V3 or more generally by Vi , where the index i takes the values 1, 2, 3. Further

1.1 Notation for Stress and Displacement

7

compression is then achieved by using the Einstein summation convention according to which terms containing a repeated latin index are summed over the three values 1, 2, 3. For example, the expression σii is interpreted as σii ≡

3 

σii = σ11 + σ22 + σ33 .

i=1

Occasionally, we may want to use a repeated index without implying a summation. In such cases, we shall add the explicit instruction ‘no sum’. Any expression with one free index has a value associated with each coördinate direction and hence represents a vector. A more formal connection between the two notations can be established by writing V = e1 V1 + e2 V2 + e3 V3 = ei Vi ,

(1.3)

where e1 = i, e2 = j , e3 = k and then dropping the implied unit vector ei in the right-hand side of (1.3). The position of a point in space is identified by the position vector, r = i x + j y + kz . In index notation, we replace x, y, z by x1 , x2 , x3 respectively, so that r = {x1 , x2 , x3 }T , which is simply written as xi , whilst the Cartesian stress components ⎡

⎤ σ11 σ12 σ13 σ = ⎣ σ21 σ22 σ23 ⎦ σ31 σ32 σ33

(1.4)

are written as σi j . Many authors use a subscript comma followed by one or more indices to denote differentiation with respect to Cartesian coördinates. Thus u j,i =

∂u j ∂xi

and

φ,ii =

∂2φ ∂2φ ∂2φ ∂2φ = + + . ∂xi ∂xi ∂x1 2 ∂x2 2 ∂x3 2

This clearly leads to a more compact notation which has advantages when writing general formulæ, but we shall not use it in this book because the explicit derivative is easier to interpret, particularly for those meeting the index notation for the first time.

8

1 Introduction

1.1.3 Vector operators in index notation All the well-known vector operations can be performed using index notation and the summation convention. For example, if two vectors are represented by Pi , Q i respectively, their scalar (dot) product can be written concisely as P · Q = P1 Q 1 + P2 Q 2 + P3 Q 3 ≡ Pi Q i ,

(1.5)

because the repeated index i in the last expression implies the summation over all three product terms. The vector (cross) product i j k P × Q = P1 P2 P3 Q1 Q2 Q3 can be written in index notation as P × Q = i jk Pi Q j ,

(1.6)

where i jk is the alternating tensor, which is defined to be 1 if the indices are in cyclic order (e.g. 2,3,1), –1 if they are in reverse cyclic order (e.g. 2,1,3) and zero if any two indices are the same3 . Notice that the only free index in the right-hand side of (1.6) is k, so (for example) the x1 -component of P × Q is recovered by setting k = 1. The gradient of a scalar function φ can be written ∇φ = i

∂φ ∂φ ∂φ ∂φ + j +k = , ∂x ∂y ∂z ∂xi

(1.7)

the divergence of a vector V is div V ≡ ∇·V =

∂V2 ∂V3 ∂Vi ∂V1 + + = ∂x1 ∂x2 ∂x3 ∂xi

using (1.5, 1.7), and the curl of a vector V is curl V ≡ ∇×V = i jk

∂V j , ∂xi

using (1.6, 1.7). We also note that the Laplacian of a scalar function can be written in any of the forms ∇2φ = 3

∂2φ ∂2φ ∂2φ ∂2φ + + 2 = div ∇φ = . 2 2 ∂x ∂y ∂z ∂xi ∂xi

In two dimensional problems, the same rules can be used to define i j , with i, j = 1, 2.

1.1 Notation for Stress and Displacement

9

1.1.4 Vectors, tensors and transformation rules Vectors can be conceived in a mathematical sense as ordered sets of numbers, or in a physical sense as mathematical representations of quantities characterized by magnitude and direction. A link between these concepts is provided by the transformation rules. Suppose we know the components (Vx , Vy ) of the vector V in a given two-dimensional Cartesian coördinate system (x, y), and we wish to determine the components (Vx , Vy ) in a new system (x  , y  ) which is inclined to (x, y) at an angle θ in the anticlockwise direction as shown in Figure 1.3.

Fig. 1.3 The coördinate systems x, y and x  , y  .

The required components are Vx = Vx cos θ + Vy sin θ ;

Vy = Vy cos θ − Vx sin θ .

(1.8)

We could define a vector as an entity, described by its components in a specified Cartesian coördinate system, which transforms into other coördinate systems according to rules like equations (1.8) — i.e. as an ordered set of numbers which obey the transformation rules (1.8). The idea of magnitude and direction could then be introduced by noting that we can always choose θ such that (i) Vy = 0 and (ii) Vx > 0. The corresponding direction x  is then the direction of the resultant vector and the component Vx is its magnitude. Now stresses have two suffices and components associated with all possible combinations of two coördinate directions, though we note that equation (1.1) shows that the order of the suffices is immaterial. (Another way of stating this is that the matrix of stress components (1.4) is always symmetric). The stress components satisfy a more complicated set of transformation rules which can be determined by considering the equilibrium of an infinitesimal wedge-shaped piece of material. The resulting equations in the two-dimensional case are those associated with Mohr’s circle — i.e. σx x = σx x cos2 θ + σ yy sin2 θ + 2σx y sin θ cos θ

 

σx y = σx y cos2 θ − sin2 θ + σ yy − σx x sin θ cos θ

(1.10)

σ yy

(1.11)

= σ yy cos θ + σx x sin θ − 2σx y sin θ cos θ , 2

2

(1.9)

10

1 Introduction

where we use the notation σx x rather than σx  x  in the interests of clarity. As in the case of vectors we can define a mathematical entity which has a matrix of components in any given Cartesian coördinate system and which transforms into other such coördinate systems according to rules like (1.9–1.11). Such quantities are called second order Cartesian tensors. Notice incidentally that the alternating tensor introduced in equation (1.6) is not a second order Cartesian tensor. We know from Mohr’s circle that we can always choose θ such that σx y = 0, in which case the directions x  , y  are referred to as principal directions and the components σx x , σ yy as principal stresses. Thus another way to characterize a second order Cartesian tensor is as a quantity defined by a set of orthogonal principal directions and a corresponding set of principal values. As with vectors, a pragmatic motivation for abstracting the mathematical properties from the physical quantities which exhibit them is that many different physical quantities are naturally represented as second order Cartesian tensors. Apart from stress and strain, some commonly occurring examples are the second moments of area of a beam cross section (Ix , Ix y , I y ), the second partial derivatives of a scalar function (∂ 2 f /∂x 2 ; ∂ 2 f /∂x∂ y; ∂ 2 f /∂ y 2 ) and the influence coefficient (compliance) matrix Ci j defining the displacement u due to a force F for a linear elastic system, i.e. u i = Ci j F j ,

(1.12)

where the summation convention is implied. It is a fairly straightforward matter to prove that each of these quantities obeys transformation rules like (1.9–1.11). It follows immediately (for example) that every beam cross section has two orthogonal principal axes of bending about which the two principal second moments are respectively the maximum and minimum for the cross section. A special tensor of some interest is that for which the Mohr’s circle degenerates to a point. In the case of stresses, this corresponds to a state of hydrostatic stress, so-called because a fluid at rest cannot support shear stress (the constitutive law for a fluid relates velocity gradient to shear stress) and hence σx y = 0 for all θ. The only Mohr’s circle which satisfies this condition is one of zero radius, from which we deduce immediately that all directions are principal directions and that the principal values are all equal. In the case of the fluid, we obtain the well-known result that the pressure in a fluid at rest is equal in all directions. It is instructive to consider this result in the context of other systems involving tensors. For example, consider the second moments of area for the square cross section shown in Figure 1.4. By symmetry, we know that O x, O y are principal directions and that the two principal second moments are both equal to a 4 /12. It follows that the Mohr’s circle of second moments has zero radius and hence (i) that the second moment about any other axis must also be a 4 /12 and (ii) that the product

1.1 Notation for Stress and Displacement

11

Fig. 1.4 A beam of square cross section.

inertia Ix y must be zero about all axes, showing that the axis of curvature is always aligned with the applied bending moment. These results are not at all obvious merely from an examination of the section. As a second example, Figure 1.5 shows an elastic system consisting of three identical but arbitrary structures connecting a point, P, to a rigid support, the structures being inclined to each other at angles of 120o .

Fig. 1.5 Support structure with three similar but unsymmetrical components.

The structures each have elastic properties expressible in the form of an influence function matrix as in equation (1.12) and are generally such that the displacement u is not colinear with the force F. However, the overall influence function matrix for the system has the same properties in three different coördinate systems inclined to each other at 120o , since a rotation of the figure through 120o leaves the system unchanged. The only Mohr’s circle which gives equal components after a rotation of 120o is that of zero radius. We therefore conclude that the support system of Figure 1.5 is such that (i) the displacement of P always has the same direction as the force F and (ii) the stiffness or compliance of the system is the same in

12

1 Introduction

all directions4 . These two examples illustrate that there is sometimes an advantage to be gained from considering a disparate physical problem that shares a common mathematical structure with that under investigation.

Three-dimensional transformation rules In three dimensions, the vector transformation rules (1.8) are most conveniently written in terms of direction cosines li j , defined as the cosine of the angle between unit vectors in the directions xi and x j respectively. We then obtain Vi = li j V j .

(1.13)

For the two-dimensional case of Figure 1.3 and equations (1.8), the matrix5 formed by the components li j takes the form

cos θ li j = − sin θ

sin θ cos θ

.

(1.14)

We already remarked in §1.1.1 that only those stress components with an xsuffix act on the surface x = c and hence that if this surface is traction free (for example if it is an unloaded boundary of a body), then σx x = σx y = σx z = 0, but no restrictions are placed on the components σ yy , σ yz , σzz . More generally, the traction t on a specified plane is a vector with three independent components ti , in contrast to the stress σi j which is a second order tensor with six independent components. Mathematically, we can define the orientation of a specified inclined plane by a unit vector n in the direction of the outward normal. By considering the equilibrium of an infinitesimal tetrahedron whose surfaces are perpendicular to the directions x1 , x2 , x3 and n respectively, we then obtain ti = σi j n j ,

(1.15)

for the traction vector on the inclined surface. The components ti defined by (1.15) are aligned with the Cartesian axes xi , but we can use the vector transformation rule (1.13) to resolve t into a new coördinate system in which one axis (x1 say) is aligned with n. We then have n j = l1 j ; ti = σi j l1 j 4

and

tk = lki ti = lki l1 j σi j .

A similar argument can be used to show that if a laminated fibre-reinforced composite is laid up with equal numbers of identical, but not necessarily symmetrical, laminæ in each of 3 or more equispaced orientations, it must be elastically isotropic within the plane. This proof depends on the properties of the fourth order Cartesian tensor ci jkl describing the stress-strain relation of the laminæ (see equations (1.40 1.43) below). 5 Notice that the matrix of direction cosines is not symmetric.

1.1 Notation for Stress and Displacement

13

If we perform this operation for each of the three surfaces orthogonal to x1 , x2 , x3 respectively, we shall recover the complete stress tensor in the rotated coördinate system, which is therefore given by σi j = li p l jq σ pq

or

σ = Ω σ Ω T ,

(1.16)

where Ω is an unsymmetric matrix whose components Ωi j = li j . Comparison of (1.16) with (1.9–1.11) provides a good illustration of the efficiency of the index notation in condensing general expressions in Cartesian coördinates.

1.1.5 Principal stresses and von Mises stress One of the principal reasons for performing elasticity calculations is to determine when and where an engineering component will fail. Theories of material failure are beyond the scope of this book, but the most widely used criteria are the von Mises distortion energy criterion for ductile materials and the maximum tensile stress criterion for brittle materials6 . Brittle materials typically fracture when the maximum tensile stress reaches a critical value and hence we need to be able to calculate the maximum principal stress σ1 . For two dimensional problems, the principal stresses can be found by using the condition σx y = 0 in equation (1.10) to determine the inclination θ of the principal directions. Substituting into (1.9, 1.11) then yields the well known results σx x + σ yy ± σ1 , σ2 = 2



σx x − σ yy 2

2 + σx2y .

In most problems, the maximum stresses occur at the boundaries where shear tractions are usually zero. Thus, even in three-dimensional problems, the determination of the maximum tensile stress often involves only a two-dimensional stress transformation. However, the principal stresses are easily obtained in the fully threedimensional case from the results   2 I1 2 (1.17) + I1 − 3I2 cos φ σ1 = 3 3     2π 2 I1 + (1.18) σ2 = I12 − 3I2 cos φ + 3 3 3     4π 2 I1 + , (1.19) σ3 = I12 − 3I2 cos φ + 3 3 3 where 6

J. R. Barber, Intermediate Mechanics of Materials, Springer, Dordrecht, 2nd. edn. (2011), §2.2.

14

1 Introduction

  3 2I1 − 9I1 I2 + 27I3 1 φ = arccos 3 2(I12 − 3I2 )3/2 and I1 = σx x + σ yy + σzz

(1.20)

I2 = σx x σ yy + σ yy σzz + σzz σx x − I3 = σx x σ yy σzz −

σx x σ 2yz

−

σx2y

2 σ yy σzx

−

2 − σ 2yz − σzx σzz σx2y + 2σx y σ yz σzx

(1.21) .

(1.22)

The quantities I1 , I2 , I3 are known as stress invariants because for a given stress state they are the same in all coördinate systems. If the principal value of θ = arccos(x) is defined such that 0 ≤ θ < π, the principal stresses defined by equations (1.17–1.19) will always satisfy the inequality σ1 ≥ σ3 ≥ σ2 . Von Mises theory states that a ductile material will yield when the strain energy of distortion per unit volume reaches a certain critical value. This enables us to define an equivalent tensile stress or von Mises stress σ E as σE ≡ =



 

I12

− 3I2 =

(3σi j σ ji − σii σ j j ) 2

2 −σ σ −σ σ −σ σ 2 2 2 σx2x + σ 2yy + σzz x x yy yy zz zz x x + 3σx y + 3σ yz + 3σzx .

(1.23) The von Mises yield criterion then states that yield will occur when σ E reaches the yield stress in uniaxial tension, SY . The Maple and Mathematica files ‘principalstresses’ use equations (1.17–1.23) to calculate the principal stresses and the von Mises stress from a given set of stress components.

1.1.6 Displacement The displacement of a particle P is a vector u = iu x + j u y + ku z representing the difference between the final and the initial position of P — i.e. it is the distance that P moves during the deformation. In index notation, the displacement is represented as u i . The deformation of a body is completely defined if we know the displacement of its every particle. Notice however that there is a class of displacements which do not involve deformation — the so-called ‘rigid-body displacements’. A typical case is where all the particles of the body have the same displacement. The name arises, of course, because rigid-body displacement is the only class of displacement that can be experienced by a rigid body.

1.2 Strains and their Relation to Displacements

15

1.2 Strains and their Relation to Displacements Components of strain will be denoted by the symbol e, with appropriate suffices (e.g. ex x , ex y ). As in the case of stress, no special symbol is required for shear strain, though we shall see below that the quantity defined in most elementary texts (and usually denoted by γ) differs from that used in the mathematical theory of Elasticity by a factor of 2. A major advantage of this definition is that it makes the strain e a second order Cartesian Tensor (see §1.1.4 above). We shall demonstrate this by establishing transformation rules for strain similar to equations (1.9–1.11) in §1.2.4 below.

1.2.1 Tensile strain Students usually first encounter the concept of strain in elementary Mechanics of Materials as the ratio of extension to original length and are sometimes confused by the apparently totally different definition used in more mathematical treatments of solid mechanics. We shall discuss here the connection between the two definitions — partly for completeness, and partly because the physical insight that can be developed in the simple problems of Mechanics of Materials is very useful if it can be carried over into more difficult problems. Figure 1.6 shows a bar of original length L and density ρ hanging from the ceiling. Suppose we are asked to find how much it increases in length under the loading of its own weight.

Fig. 1.6 The bar suspended from the ceiling.

It is easily shown that the tensile stress σx x at the point P, distance x from the ceiling, is σx x = ρg(L − x) , where g is the acceleration due to gravity, and hence from Hooke’s law,

16

1 Introduction

ex x =

ρg(L − x) . E

(1.24)

However, the strain varies continuously over the length of the bar and hence we can only apply the Mechanics of Materials definition if we examine an infinitesimal piece of the bar over which the strain can be regarded as sensibly constant. We describe the deformation in terms of the downward displacement u x which depends upon x and consider that part of the bar between x and x +δx, denoted by P Q in Figure 1.7.

Fig. 1.7 Infinitesimal section of the bar.

After the deformation, P Q must have extended by u x (Q) − u x (P) and hence the local value of ‘Mechanics of Materials’ tensile strain is ex x =

u x (x + δx) − u x (x) u x (Q) − u x (P) = . δx δx

Taking the limit as δx → 0, we obtain the definition ex x =

∂u x . ∂x

(1.25)

Corresponding definitions can be developed in three-dimensional problems for the other normal strain components, i.e. e yy =

∂u y ∂u z ; ezz = . ∂y ∂z

(1.26)

Notice how easy the problem of Figure 1.6 becomes when we use these definitions. We get ∂u x ρg(L − x) = , ∂x E from (1.24, 1.25) and hence

1.2 Strains and their Relation to Displacements

ux =

17

ρg(2L x − x 2 ) +A, 2E

(1.27)

where A is an arbitrary constant of integration which expresses the fact that our knowledge of the stresses and hence the strains in the body is not sufficient to determine its position in space. In fact, A represents an arbitrary rigid-body displacement. In this case we need to use the fact that the top of the bar is joined to a supposedly rigid ceiling. Thus, u x (0) = 0 and hence A = 0 from (1.27). The increase of length of the bar is represented by the displacement at x = L and is u x (L) =

ρgL 2 . 2E

1.2.2 Rotation and shear strain Noting that the two x’s in ex x correspond to those in its definition ∂u x /∂x, it is natural to seek a connection between the shear strain ex y and one or both of the derivatives ∂u x /∂ y, ∂u y /∂x. As a first step, we shall discuss the geometrical interpretation of these derivatives.

Fig. 1.8 Rotation of a line segment.

Figure 1.8 shows a line segment P Q of length δx, aligned with the x-axis, the two ends of which are displaced in the y-direction. Clearly if u y (Q) = u y (P), the line P Q will be rotated by these displacements and if the angle of rotation is small it can be written u y (x + δx) − u y (x) , φ= δx (anticlockwise positive). Proceeding to the limit as δx → 0, we have

18

1 Introduction

φ=

∂u y . ∂x

Thus, ∂u y /∂x is the angle through which a line originally in the x-direction rotates towards the y-direction during the deformation7 . Now, if P Q is a line drawn on the surface of an elastic solid, the occurrence of a rotation φ does not necessarily indicate that the solid is deformed — we could rotate the line simply by rotating the solid as a rigid body. To investigate this matter further, we imagine drawing a series of lines at different angles through the point P as shown in Figure 1.9(a).

Fig. 1.9 Rotation of lines at point P; (a) Original state, (b) Rigid-body rotation; (c) Rotation and deformation.

If the vicinity of the point P suffers merely a local rigid-body rotation, all the lines will rotate through the same angle and retain the same relative inclinations as shown in 1.9(b). However, if different lines rotate through different angles, as in 1.9(c), the body must have been deformed. We shall show in the next section that the rotations of the lines in Figure 1.9(c) are not independent and a consideration of their interdependency leads naturally to a definition of shear strain.

1.2.3 Transformation of coördinates Suppose we knew the displacement components u x , u y throughout the body and wished to find the rotation, φ, of the line P Q in Figure 1.10, which is inclined at an angle θ to the x-axis. We construct a new axis system O x  y  with O x  parallel to P Q as shown, in which case we can argue as above that P Q rotates anticlockwise through the angle

7

Students of Mechanics of Materials will have already used a similar result when they express the slope of a beam as du/d x, where x is the distance along the axis of the beam and u is the transverse displacement.

1.2 Strains and their Relation to Displacements

φ(P Q) =

19

∂u y ∂x 

.

(1.28)

Fig. 1.10 Rotation of a line inclined at angle θ.

Furthermore, we have   ∂ ∂ ∂ ∂ ∂  + j ·i  = i · i  + j · i =∇·i = i ∂x  ∂x ∂y ∂x ∂y ∂ ∂ + sin θ , = cos θ ∂x ∂y

(1.29)

and by similar arguments, u x = u x cos θ + u y sin θ ;

u y = u y cos θ − u x sin θ .

(1.30)

In fact equations (1.29, 1.30) are simply restatements of the vector transformation rules (1.8), since both u and the gradient operator ∇ are vectors. Substituting these results into equation (1.28), we find   ∂ ∂ + sin θ (u y cos θ − u x sin θ) φ(P Q) = cos θ ∂x ∂y   ∂u y ∂u y ∂u x ∂u x cos2 θ − sin2 θ + − sin θ cos θ = ∂x ∂y ∂y ∂x     ∂u x 1 ∂u y ∂u x 1 ∂u y − + + cos(2θ) = 2 ∂x ∂y 2 ∂x ∂y   ∂u x 1 ∂u y − sin(2θ) . (1.31) + 2 ∂y ∂x In the final expression (1.31), the first term is independent of the inclination θ of the line P Q and hence represents a rigid-body rotation as in Figure 1.9(b). We denote this rotation by the symbol ω, which with the convention illustrated in Figure 1.8 is anticlockwise positive. Notice that as this term is independent of θ in equation (1.31), it is the same for any right-handed set of axes — i.e.

20

1 Introduction

1 ω= 2



∂u y ∂u x − ∂x ∂y



1 = 2



∂u y

∂u x − ∂x  ∂ y

 ,

(1.32)

for any x  , y  . In three dimensions, ω represents a small clockwise rotation about the positive z-axis and is therefore more properly denoted by ωz to distinguish it from the corresponding rotations about the x and y axes — i.e.  ∂u y ∂u x − . ∂x ∂y (1.33) The rotation ω is therefore a vector in three-dimensional problems and is more compactly defined in vector or index notation as ωx =

1 2



∂u y ∂u z − ∂y ∂z

ω=



; ωy =

1 2



∂u x ∂u z − ∂z ∂x

1 1 curl u ≡ ∇ ×u 2 2

or



; ωz =

ωk =

1 2



  ∂u j 1 , i jk 2 ∂xi

(1.34)

where i jk is defined in §1.1.3. In two dimensions ω behaves as a scalar since two of its components degenerate to zero8 .

1.2.4 Definition of shear strain We are now in a position to define the shear strain ex y as the difference between the rotation of a line drawn in the x-direction and the corresponding rigid-body rotation, ωz , i.e.   ∂u y ∂u x 1 ∂u y − ωz = + (1.35) ex y = ∂x 2 ∂x ∂y and similarly e yz =

1 2



∂u y ∂u z + ∂y ∂z

 ; ezx =

1 2



∂u x ∂u z + ∂z ∂x

 .

(1.36)

Note that ex y so defined is one half of the quantity γx y used in Mechanics of Materials and in many older books on Elasticity. The strain-displacement relations (1.25, 1.26, 1.35, 1.36) can be written in the concise form   ∂u j 1 ∂u i . (1.37) + ei j = 2 ∂x j ∂xi

In some books, ωx , ω y , ωz are denoted by ω yz , ωzx , ωx y respectively. This notation is not used here because it gives the erroneous impression that ω is a second order tensor rather than a vector.

8

1.3 Stress-strain Relations

21

With the notation of equations (1.25, 1.26, 1.33, 1.35), we can now write (1.31) in the form φ(P Q) = ωz + ex y (cos2 θ − sin2 θ) + (e yy − ex x ) sin θ cos θ and hence ex y = φ(P Q) − ωz

(by definition)

= ex y (cos θ − sin θ) + (e yy − ex x ) sin θ cos θ . 2

2

(1.38)

This is one of the coördinate transformation relations for strain, the other one ex x = ex x cos2 θ + e yy sin2 θ + 2ex y sin θ cos θ

(1.39)

being obtainable from equations (1.25, 1.29, 1.30) in the same way. A comparison of equations (1.38, 1.39) and (1.9, 1.10) confirms that, with these definitions, the strain ei j is a second order Cartesian tensor. The corresponding three-dimensional straintransformation equations have the same form as (1.16) and can be obtained using the strain-displacement relation (1.37) and the vector transformation rule (1.13). This derivation is left as an exercise for the reader (Problem 1.9).

1.3 Stress-strain Relations The fundamental assumption of linear elasticity is that the material obeys Hooke’s law, implying that there is a linear relation between stress and strain. The most general such relation can be written in index notation as σi j = ci jkl ekl ; ei j = si jkl σkl ,

(1.40)

where ci jkl , si jkl are the elasticity tensor and the compliance tensor respectively. Both the elasticity tensor and the compliance tensor must satisfy the symmetry conditions ci jkl = c jikl = ckli j = ci jlk ,

(1.41)

which follow from (i) the symmetry of the stress and strain tensors (e.g. σi j = σ ji ) and (ii) the reciprocal theorem, which we shall discuss in Chapter 38. Substituting the strain-displacement relations (1.37) into the first of (1.40) and using (1.41) to combine appropriate terms, we obtain σi j = ci jkl

∂u k . ∂xl

(1.42)

22

1 Introduction

Equation (1.16) and the corresponding strain-transformation equation imply that the fourth-order tensors ci jkl , si jkl take the form ci jkl = li p l jq lkr lls c pqr s ; si jkl = li p l jq lkr lls s pqr s ,

(1.43)

in a coördinate system defined by the direction cosines li j . In this book, we shall mostly restrict attention to the case where the material is isotropic, meaning that the tensors ci jkl , si jkl are invariant under all such coördinate transformations, and (e.g.) ci jkl = ci jkl . In other words, a test specimen of the material would show identical behaviour regardless of its orientation relative to the original piece of material from which it was manufactured. However, the more general anisotropic case will be discussed in Chapter 36.

1.3.1 Isotropic constitutive law We shall develop the various forms of Hooke’s law for the isotropic medium starting from the results of the uniaxial tensile test, in which the only non-zero stress component is σx x . Symmetry considerations then show that the shear strains ex y , e yz , ezx must be zero and that the transverse strains e yy , ezz must be equal. Thus, the most general relation can be written in the form ex x =

σx x νσx x ; e yy = ezz = − , E E

where Young’s modulus E and Poisson’s ratio ν are experimentally determined constants. The normal strain components due to a more general triaxial state of stress can then be written down by linear superposition as νσ yy νσzz σx x − − E E E σ yy νσx x νσzz e yy = − + − E E E νσ yy σzz νσx x − + . ezz = − E E E

ex x =

(1.44) (1.45) (1.46)

The relation between ex y and σx y can then be obtained by using the coördinate transformation relations. We know that there are three principal directions such that if we align them with x, y, z, we have σx y = σ yz = σzx = 0

(1.47)

ex y = e yz = ezx = 0 .

(1.48)

and hence by symmetry

1.3 Stress-strain Relations

23

Using a system aligned with the principal directions, we write ex y = (e yy − ex x ) sin θ cos θ from equations (1.38, 1.48), and hence (σ yy − σx x )(1 + ν) sin θ cos θ E (1 + ν)σx y , = E

ex y =

(1.49)

from (1.44, 1.45, 1.10, 1.47). We define the shear modulus or the modulus of rigidity μ=

E , 2(1 + ν)

so that equation (1.49) takes the form ex y =

σx y 2μ

.

(1.50)

1.3.2 Lamé’s constants It is often desirable to solve equations (1.44–1.46) to express σx x in terms of ex x etc. The solution is routine and leads to the equation σx x =

Eν(ex x + e yy + ezz ) Eex x + (1 + ν)(1 − 2ν) (1 + ν)

and similar equations, which are more concisely written in the form σx x = λe + 2μex x etc., where

(1.51)

2μν Eν = (1 + ν)(1 − 2ν) (1 − 2ν)

(1.52)

e ≡ ex x + e yy + ezz ≡ eii ≡ div u

(1.53)

λ= and

is known as the dilatation. The stress-strain relations (1.50, 1.51) can be written more concisely in the index notation in the form (1.54) σi j = λδi j emm + 2μei j ,

24

1 Introduction

where δi j is the Kronecker delta, defined as 1 if i = j and 0 if i = j. Equivalently, we can use equation (1.40) with 

ci jkl = λδi j δkl + μ δik δ jl + δ jk δil .

(1.55)

The constants λ, μ are known as Lamé’s constants. Young’s modulus and Poisson’s ratio can be written in terms of Lamé’s constants through the equations E=

μ(3λ + 2μ) ; (λ + μ)

ν=

λ . 2(λ + μ)

1.3.3 Dilatation and bulk modulus The dilatation, e, is easily shown to be invariant as to coördinate transformation and is therefore a scalar quantity. In physical terms it is the local volumetric strain, since a unit cube increases under strain to a block of dimensions (1+ex x ), (1+e yy ), (1+ezz ) and hence the volume change is ΔV = (1 + ex x )(1 + e yy )(1 + ezz ) − 1 = ex x + e yy + ezz + O(ex x e yy ) . It can be shown9 that the dilatation e and the rotation vector ω are harmonic — i.e. ∇ e = ∇ 2 ω = 0. For this reason, many early solutions of elasticity problems were formulated in terms of these variables, so as to make use of the wealth of mathematical knowledge about harmonic functions. We now have other more convenient ways of expressing elasticity problems in terms of harmonic functions, which will be discussed in Chapter 21 et seq.. From equations (1.44–1.46), we have 2

e = eii =

(σx x + σ yy + σzz )(1 − 2ν) σ = , E Kb

where σ=

σx x + σ yy + σzz 1 = σii 3 3

(1.56)

is the mean normal stress, also known as the bulk stress or the hydrostatic stress, and Kb =

2μ(1 + ν) 2μ E = =λ+ 3(1 − 2ν) 3(1 − 2ν) 3

(1.57)

is the bulk modulus. We note that K b → ∞ if ν → 0.5, in which case the material is described as incompressible. 9

See Problems 2.1, 2.2.

Problems

25

1.3.4 Deviatoric stress It is sometimes convenient to represent a general state of stress σi j as the sum of a hydrostatic stress σ and a deviatoric stress σi j defined as σi j = σi j − σδi j .

(1.58)

We also define the deviatoric strain ei j as e ei j = ei j − δi j , 3

(1.59)

in which case it is easily shown that σi j = 2μei j . In particular, the von Mises stress of equation (1.23) depends only on the deviatoric stress. In other words, if equation (1.58) is used to eliminate σi j in (1.23), the terms involving σ cancel.

Problems 1.1. Show that (i)

∂xi = δi j ∂x j

and

(ii) R =

√

xi xi ,

where R = |r| is the distance from the origin. Hence find ∂ R/∂x j in index notation. Confirm your result by finding ∂ R/∂x in x, y, z notation. 1.2. Prove that the partial derivatives ∂ 2 f /∂x 2 , ∂ 2 f /∂x∂ y, ∂ 2 f /∂ y 2 of the scalar function f (x, y) transform into the rotated coördinate system x  , y  by rules similar to equations (1.9–1.11). 1.3. Show that the direction cosines defined in (1.13) satisfy the identity li j lik = δ jk . Hence or otherwise, show that the product σi j σi j is invariant under coördinate transformation. 1.4. By restricting the indices i, j etc. to the values 1,2 only, show that the twodimensional stress-transformation relations (1.9–1.11) can be obtained from (1.16) using the two-dimensional direction cosines (1.14). 1.5. Use the index notation to develop concise expressions for the three stress invariants I1 , I2 , I3 .

26

1 Introduction

1.6. Choosing a local coördinate system x1 , x2 , x3 aligned with the three principal axes, determine the tractions on the octahedral plane defined by the unit vector  n=

1 1 1 √ ,√ ,√ 3 3 3

T

which makes equal angles with all three principal axes, if the principal stresses are , σ2 , σ3 . Hence show that the magnitude of the resultant shear stress on this plane σ1√ is 2σ E /3, where σ E is given by equation (1.23). 1.7. A rigid body is subjected to a small rotation ωz = Ω  1 about the z-axis. If the displacement of the origin is zero, find expressions for the three displacement components u x , u y , u z as functions of x, y, z. 1.8. Use the index notation to develop a general expression for the derivative ∂u i ∂x j in terms of strains and rotations. 1.9. Use the three-dimensional vector transformation rule (1.13) and the index notation to prove that the strain components (1.37) transform according to the equation ei j = li p l jq e pq . Hence show that the dilatation eii is invariant under coördinate transformation. 1.10. Find an index notation expression for the compliance tensor si jkl of equation (1.40) for the isotropic elastic material in terms of the elastic constants E, ν. 1.11. Show that equations (1.44–1.46, 1.49) can be written in the concise form ei j =

νδi j σmm (1 + ν)σi j − . E E

(1.60)

Chapter 2

Equilibrium and Compatibility

We can think of an elastic solid as a highly redundant framework — each particle is built-in to its neighbours. For such a framework, we expect to get some equations from considerations of equilibrium, but not as many as there are unknowns. The deficit is made up by compatibility conditions — statements that the deformed components must fit together. These latter conditions will relate the dimensions and hence the strains of the deformed components and in order to express them in terms of the same unknowns as the stresses (forces) we need to make use of the stress-strain relations as applied to each component separately. If we were to approximate the continuous elastic body by a system of interconnected elastic bars, this would be an exact description of the solution procedure. The only difference in treating the continuous medium is that the system of algebraic equations is replaced by partial differential equations describing the same physical or geometrical principles.

2.1 Equilibrium Equations We consider a small rectangular block of material — side δx, δ y, δz — as shown in Figure 2.1. We suppose that there is a body force1 , p per unit volume, and that the stresses vary with position so that the stress components on opposite faces of the block differ by the differential quantities δσx x , δσx y etc. Only those stress components which act in the x-direction are shown in Figure 2.1 for clarity.

1

A body force is one that acts directly on every particle of the body, rather than being applied by tractions at its boundaries and transmitted to the various particles by means of internal stresses. This is an important distinction which will be discussed further in Chapter 7, below. The commonest example of a body force is that due to gravity.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 J. R. Barber, Elasticity, Solid Mechanics and Its Applications 172, https://doi.org/10.1007/978-3-031-15214-6_2

27

28

2 Equilibrium and Compatibility

Fig. 2.1 Forces in the x-direction on an elemental block.

Resolving forces in the x-direction, we find (σx x + δσx x − σx x )δ yδz + (σx y + δσx y − σx y )δzδx +(σx z + δσx z − σx z )δxδ y + px δxδ yδz = 0 . Hence, dividing through by (δxδ yδz) and proceeding to the limit as these infinitesimals tend to zero, we obtain ∂σx y ∂σx z ∂σx x + + + px = 0 . ∂x ∂y ∂z

(2.1)

Similarly, we have ∂σ yx ∂σ yy ∂σ yz + + + py = 0 ∂x ∂y ∂z ∂σzy ∂σzz ∂σzx + + + pz = 0 , ∂x ∂y ∂z or in index notation

∂σi j + pi = 0 . ∂x j

(2.2) (2.3)

(2.4)

These are the differential equations of equilibrium.

2.2 Compatibility Equations The easiest way to satisfy the equations of compatibility — as in framework problems — is to express all the strains in terms of the displacements. In a framework, this ensures that the components fit together by identifying the displacement of points in two links which are pinned together by the same symbol. If the framework is redundant, the number of pin displacements thereby introduced is less than the number of component lengths (and hence extensions) determined by them — and the number of unknowns is therefore reduced.

2.2 Compatibility Equations

29

For the continuum, the process is essentially similar, but much more straightforward. We define the six components of strain in terms of displacements through equations (1.37). These six equations introduce only three unknowns (u x , u y , u z ) and hence the latter can be eliminated to give equations constraining the strain components. For example, from (1.25, 1.26) we find ∂ 2 e yy ∂3u y ∂3u x ∂ 2 ex x = ; = ∂ y2 ∂x∂ y 2 ∂x 2 ∂ y∂x 2 and hence

∂ 2 e yy ∂2 ∂ 2 ex x + = ∂ y2 ∂x 2 ∂x∂ y

from (1.35) — i.e.



∂u y ∂u x + ∂y ∂x

 =2

∂ 2 ex y , ∂x∂ y

∂ 2 e yy ∂ 2 ex y ∂ 2 ex x + − 2 =0. ∂ y2 ∂x∂ y ∂x 2

(2.5)

Two more equations of the same form may be obtained by permuting suffices2 . It is tempting to pursue an analogy with algebraic equations and argue that, since the six strain components are defined in terms of three independent displacement components, we must be able to develop three (i.e. 6 − 3) independent compatibility equations. However, it is easily verified that, in addition to the three equations similar to (2.5), the strains must satisfy three more equations of the form ∂ ∂ 2 ezz = ∂x∂ y ∂z



∂e yz ∂ex y ∂ezx + − ∂x ∂y ∂z

 .

(2.6)

The resulting six equations are independent in the sense that no one of them can be derived from the other five, which all goes to show that arguments for algebraic equations do not always carry over to partial differential equations. A concise statement of the six compatibility equations can be written in the index notation in the form ∂ 2 e jq =0. (2.7) i jk  pqr ∂xk ∂xr The full set of six equations makes the problem very complicated. In practice, therefore, most three-dimensional problems are treated in terms of displacements instead of strains, which satisfies the requirement of compatibility identically. However, in two dimensions, all except one of the compatibility equations degenerate to identities, so that a formulation in terms of stresses or strains is more practical.

2

i.e. x → y, y → z, z → x.

30

2 Equilibrium and Compatibility

2.2.1 The significance of the compatibility equations The physical meaning of equilibrium is fairly straightforward, but people often get mixed up about just what is being guaranteed by the compatibility equations.

Single-valued displacements Mathematically, we might say that the strains are compatible when they are definable in terms of a single-valued, continuously differentiable displacement. We could imagine reversing this process — i.e. integrating the strains (displacement gradients) to find the relative displacement (u B −u A ) of two points A, B in the solid (see Figure 2.2).

Fig. 2.2 Path of the integral in equation (2.8).

Formally we can write



B

uB − uA = A

∂u ds , ∂s

(2.8)

where s is a coördinate representing distance along a path S between A and B. If the displacements are to be single-valued, it mustn’t make any difference if we change the path S provided it remains within the solid. In other words, the integral should be path-independent. The compatibility equations are sufficient to guarantee this only if the body is simply-connected. Suppose tentatively that the contour integral 

∂u ds = 0 ∂s

(2.9)

around any infinitesimal closed loop3 . We could then make infinitesimal changes in our line from A to B by taking in such small loops until the whole line was sensibly

3

Earlier editions of this book made the erroneous claim that the compatibility equations are sufficient to guarantee this. The author is grateful to Professor Arash Yavari for drawing his attention to this error.

2.2 Compatibility Equations

31

changed, thus satisfying the requirement that the integral (2.8) is path-independent. However, the fact that the line is changed infinitesimally (continuously) prevents us from taking a qualitatively (topologically) different route through a multiplyconnected body. For example, in Figure 2.3, it is impossible to move S1 to S2 by continuous changes without passing outside the body4 . The equivalence of these two paths must therefore be enforced explicitly by requiring that  S1

or equivalently

∂u ds = ∂s  S0

 S2

∂u ds , ∂s

∂u ds = 0 , ∂s

(2.10)

where S0 = S1 − S2 (i.e. out along S1 and back along S2 ) is a closed path that encircles the hole in Figure 2.3.

Fig. 2.3 Topologically different integration paths S1 , S2 between points A and B in a multiply-connected body.

However, the compatibility equations are not quite sufficient even to guarantee (2.9). Using equations (1.34, 1.37), we can write the displacement gradient ∂u i = ei j − i jk ωk , ∂x j

(2.11)

from which we conclude that the rotation ω must also be single-valued if the derivatives in (2.8, 2.9) are to be well-defined. The rotation gradients can be expressed in terms of strain gradients through the relations

4

For a more rigorous discussion of this question, see A. E. H. Love, A Treatise on the Mathematical Theory of Elasticity, 4th edn., Dover, 1944, §156A, B. A. Boley and J. H. Weiner, Theory of Thermal Stresses, John Wiley, New York, 1960, §§3.6–3.8, or A. Yavari (2013), Compatibility equations of nonlinear elasticity for non-simply-connected bodies, Archive for Rational Mechanics and Analysis, Vol. 209, pp. 237–253.

32

2 Equilibrium and Compatibility

∂ei j ∂ωk =  jkl , ∂xi ∂xl

(2.12)

but these will integrate to a single-valued expression for ω in the body of Figure 2.3 only if    ∂ei j ∂ω ∂ωk ds ≡ d xi =  jkl d xi = 0 . (2.13) ∂s ∂x ∂xl i S0 S0 S0 If this condition is satisfied, equation (2.10) can also be expressed in terms of strains. Substituting (2.11) into (2.8) and integrating the second term by parts, we obtain 

∂u i dx j = ∂x j



 ei j d x j −

i jk ωk d x j   ∂ωk = ei j d x j − i pk x p ωk + i pk x p dx j . ∂x j

If the integral is performed around a closed path S0 , and if ω is single-valued (i.e. (2.13) is satisfied), the second term will take the same value at each end of the path and will therefore sum to zero. Substituting for the rotation gradients from (2.12) and using the identity i jk imn = δ jm δkn − δ jn δkm , the remaining terms can then be written      ∂e pj ∂ei j ∂u i ei j + x p dx j . dx j = − ∂xi ∂x p S0 ∂x j S0 A similar argument can be used to show that  u i (B) = u i (A) − i pk ωk (A) x p (B) − x p (A)    B  ∂e pj ∂ei j ei j + x p − x p (B) dx j , + − ∂xi ∂x p A again subject to the restriction that ω be single-valued5 . These results were obtained by V. Volterra and E. Cesàro and are known as Volterra-Cesàro integrals. In practice, if a solid is multiply connected it is usually easier to work in terms of displacements and by-pass this problem. Otherwise the equivalence of topologically different paths in integrals like (2.8, 2.13) must be explicitly enforced.

Compatibility of deformed shapes A more ‘physical’ way of thinking of compatibility is to state that the separate particles of the body must deform under load in such a way that they fit together 5

Notice that the first two terms in this equation correspond to rigid-body translation and rotation, so they can be set to zero if the intention is to find a particular solution for the displacements corresponding to a given strain field.

2.2 Compatibility Equations

33

after deformation. This interpretation is conveniently explored by way of a ‘jig-saw’ analogy. We consider a multiply-connected two-dimensional body cut up as a jig-saw puzzle (Figure 2.4), of which the pieces are deformable. Figure 2.4(a) shows the original puzzle and 2.4(b) the puzzle after deformation by some external loads Fi . (The pieces are shown as initially rectangular to aid visualization.) The deformation of the puzzle must satisfy the following conditions:(i) The forces on any given piece (including external forces if any) must be in equilibrium. (ii) The deformed pieces must be the right shape to fit together to make the deformed puzzle (Figure 2.4(b)).

Fig. 2.4 A multiply-connected, deformable jig-saw puzzle: (a) Before deformation; (b) After deformation; (c) Result of attempting to assemble the deformed puzzle if equation (2.9) is not satisfied for any closed path encircling the hole.

For a simply-connected body, the compatibility conditions (2.7) guarantee that (ii) is satisfied, since the deformed shape is defined by the strain components, and each piece can be given whatever rigid-body rotation is required to ensure that it fits. However, if the puzzle is multiply connected — e.g. if it has a central hole —

34

2 Equilibrium and Compatibility

condition (ii) is not sufficient to ensure that the deformed pieces can be assembled into a coherent body. Suppose we imagine assembling the ‘deformed’ puzzle working from one piece outwards. The partially completed puzzle is initially simply connected and the shape condition is sufficient to ensure the success of our assembly until we reach a piece which would convert the partially completed puzzle to a multiply-connected body. This critical piece will be the right shape for both sides, but it may be the wrong size for the separation. Also, the rigid-body rotation needed for it to fit on one side may differ from that required on the other side. If so, it will be possible to leave the completed puzzle in a state with a discontinuity as shown in Figure 2.4(c) at any arbitrarily chosen position6 , but there will be no way in which Figure 2.4(b) can be constructed. The lines defining the two sides of the discontinuity in Figure 2.4(c) are the same shape and hence, in the most general three-dimensional case, the discontinuity can be defined by six arbitrary constants corresponding to the three rigid-body translations and three rotations needed to move one side to coincide with the other. We therefore get six additional algebraic conditions for each hole in a multiply-connected body. These conditions are of course equivalent to the two vector equations (2.10, 2.13). In the two-dimensional case, the rotation degenerates to a scalar ω and there are only three integral conditions: one from equation (2.13) and two from (2.10)

2.3 Equilibrium Equations in terms of Displacements Recalling that it is often easier to work in terms of displacements rather than stresses to avoid complications with the compatibility conditions, it is convenient to express the equilibrium equations in terms of displacements. This is most easily done in index notation. Substituting for the stresses from (1.42) into the equilibrium equation (2.4), we obtain ∂2uk + pi = 0 . (2.14) ci jkl ∂x j ∂xl For the special case of isotropy, we can substitute for ci jkl from (1.55), obtaining

∂ 2 u k  λδi j δkl + μ δik δ jl + δ jk δil + pi = 0 ∂x j ∂xl

or λ

6

∂2ui ∂2uk ∂ 2 ul +μ +μ + pi = 0 . ∂xi ∂xl ∂xl ∂xl ∂xk ∂xi

since we could move one or more pieces from one side of the discontinuity to the other.

Problems

35

The first and third terms in this equation have the same form, since k, l are dummy indices. We can therefore combine them and write (λ + μ) Also, noting that

∂2ui ∂2uk +μ + pi = 0 . ∂xi ∂xk ∂xk ∂xk 

2ν λ+μ=μ 1+ 1 − 2ν

 =

μ (1 − 2ν)

from (1.52), we have ∂2uk ∂2ui (1 − 2ν) pi =0, + (1 − 2ν) + ∂xi ∂xk ∂xk ∂xk μ

(2.15)

or in vector notation ∇div u + (1 − 2ν)∇ 2 u +

(1 − 2ν) p =0. μ

(2.16)

Problems 2.1. Show that, if there are no body forces, the dilatation e must satisfy the condition ∇2e = 0 . 2.2. Show that, if there are no body forces, the rotation ω must satisfy the condition ∇2ω = 0 . 2.3. One way of satisfying the compatibility equations in the absence of rotation is to define the components of displacement in terms of a potential function ψ through the relations ∂ψ ∂ψ ∂ψ ; uy = ; uz = . ux = ∂x ∂y ∂z Use the stress-strain relations to derive expressions for the stress components in terms of ψ. Hence show that the stresses will satisfy the equilibrium equations in the absence of body forces if and only if ∇ 2 ψ = constant .

36

2 Equilibrium and Compatibility

2.4. Plastic deformation during a manufacturing process generates a state of residual stress in the large body z > 0. If the residual stresses are functions of z only and the surface z = 0 is not loaded, show that the stress components σ yz , σzx , σzz must be zero everywhere. 2.5. By considering the equilibrium of a small element of material similar to that shown in Figure 1.2, derive the three equations of equilibrium in cylindrical polar coördinates r, θ, z. 2.6. In cylindrical polar coördinates, the strain-displacement relations for the ‘inplane’ strains are err =

∂u r 1 ; er θ = ∂r 2



1 ∂u r ∂u θ uθ + − r ∂θ ∂r r

 ; eθθ =

1 ∂u θ ur + . r r ∂θ

Use these relations to obtain a compatibility equation that must be satisfied by the three strains. 2.7. If no stresses occur in a body, an increase in temperature T causes unrestrained thermal expansion defined by the strains ex x = e yy = ezz = αT ; ex y = e yz = ezx = 0 . Show that this is possible only if T is a linear function of x, y, z and that otherwise stresses must be induced in the body, regardless of the boundary conditions. 2.8. If there are no body forces, show that the equations of equilibrium and compatibility imply that ∂ 2 σi j ∂ 2 σkk + =0. (1 + ν) ∂xk ∂xk ∂xi ∂x j 2.9. Using the strain-displacement relation (1.37), show that an alternative statement of the compatibility condition is that the tensor Ci jkl ≡

∂ 2 ei j ∂ 2 e jk ∂ 2 ekl ∂ 2 eil + − − =0. ∂xk ∂xl ∂xi ∂x j ∂x j ∂xk ∂xi ∂xl

2.10. Show that the derivatives ∂ωi /∂x j needed for the evaluation of the integral (2.13) can be expressed as derivatives of the strain components ekl and find these expressions. 2.11. By differentiating equation (1.32), show that ∂ex y ∂ex x ∂ω = − ; ∂x ∂x ∂y

∂e yy ∂ex y ∂ω = − . ∂y ∂x ∂y

A particular two-dimensional strain distribution is defined by

Problems

37

ex x =

C y2 Cxy Cx2 ; e ; e . = − = x y yy (x 2 + y 2 ) (x 2 + y 2 ) (x 2 + y 2 )

Verify that these components satisfy the compatibility equation (2.5) except possibly at the origin. Then use the above results to evaluate the contour integral (2.13), using a square path S0 with corners at the four points (±1, ±1). Comment on your results. 2.12. Show that if there are no body forces, the maximum magnitude |ω| for the rotation vector must occur at a point on the boundary. 2.13. A semi-infinite prismatic body of uniform density ρ is defined by the inequalities f (x, y) < 0, z < 0, where the z-axis is vertically upwards and f is any given function of x, y only. If the surfaces of the body are all traction free, show that the state of stress is defined by σzz = ρgz, all the other stress components being zero. Show that this result remains true if the body is multiply connected. 2.14. The constitutive law (1.51) is ill-defined in the incompressible limit ν = 0.5, since λ is then unbounded and the dilatation e must be zero. An alternative formulation for this case is to express the stress components in the form ¯ ij , σi j = σi j + σδ where σi j , σ¯ are respectively the deviatoric stress and the hydrostatic stress, defined in §§1.3.3, 1.3.4. Use this formulation to develop an alternative to equation (2.16) for an incompressible material, and hence show that ∇ 2 σ¯ = −div p.

Part II

Two-dimensional Problems

Chapter 3

Plane Strain and Plane Stress

A problem is two-dimensional if the field quantities such as stress and displacement depend on only two coördinates (x, y) and the boundary conditions are imposed on a line f (x, y) = 0 in the x y-plane. In this sense, there are strictly no two-dimensional problems in elasticity. There are circumstances in which the stresses are independent of the z-coördinate, but all real bodies must have some bounding surfaces which are not represented by a line in the x y-plane. The two-dimensionality of the resulting fields depends upon the boundary conditions on such surfaces being of an appropriate form.

3.1 Plane Strain It might be argued that a closed line in the x y-plane does define a solid body — namely an infinite cylinder of the appropriate cross section whose axis is parallel to the z-direction. However, making a body infinite does not really dispose of the question of boundary conditions, since there are usually some implied boundary conditions ‘at infinity’. For example, the infinite cylinder could be in a state of uniaxial tension, σzz = C, where C is an arbitrary constant. However, a unique twodimensional infinite cylinder problem can be defined by demanding that u x , u y be independent of z and that u z = 0 for all x, y, z, in which case it follows that ezx = ezy = ezz = 0 .

(3.1)

This is the two-dimensional state known as plane strain. In view of the stress-strain relations, an equivalent statement to equation (3.1) is σzx = σzy = 0 ; u z = 0 , © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 J. R. Barber, Elasticity, Solid Mechanics and Its Applications 172, https://doi.org/10.1007/978-3-031-15214-6_3

41

42

3 Plane Strain and Plane Stress

and hence a condition of plane strain will exist in a finite cylinder provided that (i) any tractions or displacements imposed on the sides of the cylinder are independent of z and have no component in the z-direction and (ii) the cylinder has plane ends (e.g. z = ± c) which are in frictionless contact with two plane rigid walls. From the condition ezz = 0 (3.1) and the stress-strain relations, we can deduce 0=

ν(σx x + σ yy ) σzz − E E

— i.e. σzz = ν(σx x + σ yy )

(3.2)

and hence νσ yy ν 2 (σx x + σ yy ) σx x − − E E E ν(1 + ν)σ yy (1 − ν 2 )σx x = − . E E

ex x =

(3.3)

Of course, there are comparatively few practical applications in which a cylinder with plane ends is constrained between frictionless rigid walls, but fortunately the plane strain solution can be used in an approximate sense for a cylinder with any end conditions, provided that the length of the cylinder is large compared with its cross-sectional dimensions. We shall illustrate this with reference to the long cylinder of Figure 3.1, for which the ends, z = ± c are traction free and the sides are loaded by tractions which are independent of z.

Fig. 3.1 The long cylinder with traction-free ends.

We first solve the problem under the plane strain assumption, obtaining an exact solution in which all the stresses are independent of z and in which there

3.1 Plane Strain

43

exists a normal stress σzz on all z-planes, which we can calculate from equation (3.2). The plane strain solution satisfies all the boundary conditions of the problem except that σzz also acts on the end faces, z = ± c, where it appears as an unwanted normal traction. We therefore seek a corrective solution which, when superposed on the plane strain solution, removes the unwanted normal tractions on the surfaces z = ± c, without changing the boundary conditions on the sides of the cylinder.

3.1.1 The corrective solution The corrective solution must have zero tractions on the sides of the cylinder and a prescribed normal traction (equal and opposite to that obtained in the plane strain solution) on each of the end faces, A. This is a fully three-dimensional problem which generally has no closed-form solution. However, if the prescribed tractions have the linear form σzz = B + C x + Dy ,

(3.4)

the solution can be obtained in the context of Mechanics of Materials by treating the long cylinder as a beam subjected to an axial force  F=

A

σzz d xd y

(3.5)

and bending moments  Mx =

 A

σzz y d xdy ; M y = −

A

σzz x d xdy

(3.6)

about the axes O x, O y respectively, where the origin O is chosen to coincide with the centroid of the cross section1 . Notice that this correction will generally cause the cylinder to be curved, even though the bending moments in the final state are everywhere zero. If the original plane strain solution does not give a distribution of σzz of this convenient linear form, we can still use equations (3.5, 3.6) to define the force and moments for an approximate Mechanics of Materials corrective solution. The error involved in using this approximate solution will be that associated with yet another corrective solution corresponding to the problem in which the end faces of the cylinder are loaded by tractions equal to the difference between those in the plane strain 1

The Mechanics of Materials solution for axial force and pure bending is in fact exact in the sense of the Theory of Elasticity, since the stresses clearly satisfy the equilibrium equations (2.4) and the resulting strains are linear functions of x, y and hence satisfy the compatibility equations (2.7) identically.

44

3 Plane Strain and Plane Stress

solution and the Mechanics of Materials linear form (3.4) associated with the force resultants (3.5, 3.6). In this final corrective solution, the ends of the cylinder are loaded by selfequilibrated tractions, since the Mechanics of Materials approximation is carefully chosen to have the same force and moment resultants as the required exact solution and in such cases, we anticipate that significant stresses will only be generated in the immediate vicinity of the ends — or more precisely, in regions whose distance from the ends is comparable with the cross-sectional dimensions of the cylinder. If the cylinder is many times longer than its cross-sectional dimensions, there will be a substantial portion near the centre where the final corrective solution gives negligible stresses and hence where the sum of the original plane strain solution and the Mechanics of Materials correction is a good approximation to the actual three-dimensional stress field.

3.1.2 Saint-Venant’s principle The thesis that a self-equilibrating system of loads produces only local effects is known as Saint-Venant’s principle. It seems intuitively reasonable, but has not been proved rigorously except for certain special cases — some of which we shall encounter later in this book (see for example Chapter 6). It can be seen as a consequence of the rule that alternate load paths through a structure share the load in proportion with their stiffnesses. If a region of the boundary is loaded by a selfequilibrating system of tractions, the stiffest paths are the shortest — i.e. those which do not penetrate far from the loaded region. Hence, the longer paths — which are those which contribute to stresses distant from the loaded region — carry relatively little load. Note that if the local tractions are not self-equilibrating, some of the load paths must go to other distant parts of the boundary and hence there will be significant stresses in intermediate regions. For this reason, it is important to superpose the Mechanics of Materials approximate corrective solution when solving plane strain problems for long cylinders with traction-free ends. By contrast, the final stage of solving the ‘Saint-Venant problem’ to calculate the correct stresses near the ends is seldom of much importance in practical problems, since the ends, being traction free, are not generally points of such high stress as the interior. As a point of terminology, we shall refer to problems in which the boundary conditions on the ends have been corrected only in the sense of force and moment resultants as being solved in the weak sense with respect to these boundaries. Boundary conditions are satisfied in the strong sense when the tractions are specified in a pointwise rather than a force resultant sense.

3.2 Plane Stress

45

3.2 Plane Stress Plane stress is an approximate solution, in contrast to plane strain, which is exact. In other words, plane strain is a special solution of the complete three-dimensional equations of elasticity, whereas plane stress is only approached in the limit as the thickness of the loaded body tends to zero. It is argued that if the two bounding z-planes of a thin plate are sufficiently close in comparison with the other dimensions, and if they are also free of tractions, the stresses on all parallel z-planes will be sufficiently small to be neglected, in which case we write σzx = σzy = σzz = 0 , for all x, y, z. It then follows that

ezx = ezy = 0 ,

but ezz = 0, being given in fact by ezz = −

ν (σx x + σ yy ) . E

The two-dimensional stress-strain relations are then νσ yy σx x ex x = − E E σ yy νσx x − . e yy = E E

(3.7)

(3.8) (3.9)

The fact that plane stress is not an exact solution can best be explained by considering the compatibility equation ∂ 2 e yy ∂ 2 e yz ∂ 2 ezz + − 2 =0. ∂z 2 ∂ y∂z ∂ y2 Since ex-hypothesi none of the stresses vary with z, the first two terms in this equation are identically zero and hence the equation will be satisfied if and only if ∂ 2 ezz =0. ∂ y2 This in turn requires

∂ 2 (σx x + σ yy ) =0, ∂ y2

from equation (3.7), unless ν = 0, in which case the plane stress and plane strain solutions are identical. Applying similar arguments to the other compatibility equations, we conclude that for ν = 0, the plane stress assumption is exact if and only if (σx x +σ yy ) is a

46

3 Plane Strain and Plane Stress

linear function of x, y. i.e. if σx x + σ yy = B + C x + Dy . The attentive reader will notice that this condition is exactly equivalent to (3.2, 3.4). In other words, the plane stress solution is exact if and only if the ‘weak form’ solution of the corresponding plane strain solution is exact. Of course this is not a coincidence. In this special case, the process of off-loading the ends of the long cylinder in the plane strain solution has the effect of making σzz zero throughout the cylinder and hence the resulting solution also satisfies the plane stress assumption, whilst remaining exact.

3.2.1 Generalized plane stress The approximate nature of the plane stress formulation is distasteful to elasticians of a more mathematical temperament, who prefer to preserve the rigour of an exact theory. This can be done by the contrivance of defining the average stresses across the thickness of the plate — e.g. σx x ≡

1 2c



c

−c

σx x d x .

It can then be shown that the average stresses so defined satisfy the plane stress equations exactly. This is referred to as the generalized plane stress formulation. In practice, of course, the gain in rigour is illusory unless we can also establish that the stress variation across the section is small, so that the local values are reasonably close to the average. A fully three-dimensional theory of thin plates under in-plane loading shows that the plane stress assumption is a good approximation except in regions whose distance from the boundary is comparable with the plate thickness2 .

3.2.2 Relationship between plane stress and plane strain The solution of a problem under either the plane strain or plane stress assumptions involves finding a two-dimensional stress field, defined in terms of the components σx x , σx y , σ yy , which satisfies the equilibrium equations (2.4), and for which the corresponding strains, ex x , ex y , e yy , satisfy the only non-trivial compatibility equation (2.5). The equilibrium and compatibility equations are the same in both formulations,

2

See Chapter 16.

Problems

47

the only difference being in the relation between the stress and strain components, which for normal stresses are given by (3.3) for plane strain and (3.8, 3.9) for plane stress. The relation between the shear stress σx y and the shear strain ex y is the same for both formulations and is given by equation (1.50). Thus, from a mathematical perspective, the plane strain solution simply looks like the plane stress solution for a material with different elastic constants and vice versa. In fact, it is easily verified that equation (3.3) can be obtained from (3.8) by making the substitutions E=

E ν ; ν = , (1 − ν  2 ) (1 − ν  )

(3.10)

and then dropping the primes. This substitution also leaves the shear stress-shear strain relation unchanged as required. An alternative approach is to write the two-dimensional stress-strain relations in the form         κ+1 3−κ κ+1 3−κ σx x − σ yy ; e yy = σ yy − σx x ex x = 8μ 8μ 8μ 8μ σx y , ex y = 2μ where κ is Kolosov’s constant, defined as κ = (3 − 4ν)   3−ν = 1+ν

for plane strain for plane stress.

(3.11)

In the following chapters, we shall generally treat two-dimensional problems with the plane stress assumptions, noting that results for plane strain can be recovered by the substitution (3.10) when required.

Problems 3.1. The plane strain solution for the stresses in the rectangular block 0 < x < a, −b < y < b, −c < z < c with a given loading is σx x =

3F x y 3F(b2 − y 2 ) 3ν F x y ; σx y = ; σ yy = 0 ; σzz = . 3 2b 4b3 2b3

Find the tractions on the surfaces of the block and illustrate the results on a sketch of the block. We wish to use this solution to solve the corresponding problem in which the surfaces z = ± c are traction free. Determine an approximate corrective solution for this problem by offloading the unwanted force and moment resultants using the

48

3 Plane Strain and Plane Stress

elementary bending theory. Find the maximum error in the stress σzz in the corrected solution and compare it with the maximum tensile stress in the plane strain solution. 3.2. For a solid in a state of plane stress, show that if there are body forces px , p y per unit volume in the direction of the axes x, y respectively, the compatibility equation can be expressed in the form 

∇ (σx x 2

∂ py ∂ px + + σ yy ) = −(1 + ν) ∂x ∂y

 .

Hence deduce that the stress distribution for any particular case is independent of the material constants and the body forces, provided the latter are constant. 3.3. (i) Show that the compatibility equation (2.5) is satisfied by unrestrained thermal expansion (ex x = e yy = αT, ex y = 0), provided that the temperature, T , is a twodimensional harmonic function — i.e. ∂2 T ∂2 T + =0. 2 ∂x ∂ y2 (ii) Hence deduce that, subject to certain restrictions which you should explicitly list, no thermal stresses will be induced in a thin body with a steady-state, twodimensional temperature distribution and no boundary tractions. (iii) Show that an initially straight line on such a body will be distorted by the heat flow in such a way that its curvature is proportional to the local heat flux across it. 3.4. Find the inverse relations to equations (3.10) — i.e. the substitutions that should be made for the elastic constants E, ν in a plane strain solution if we want to recover the solution of the corresponding plane stress problem. 3.5. Show that in a state of plane stress without body forces, the in-plane displacements must satisfy the equations ∇2u x +

        1+ν ∂ ∂u x ∂u y 1 + ν ∂ ∂u x ∂u y + = 0 ; ∇2u y + + =0. 1−ν ∂x ∂x ∂y 1 − ν ∂ y ∂x ∂y

3.6. Show that in a state of plane strain without body forces, ∂e = ∂x



1 − 2ν 1−ν



  ∂ωz ∂e 1 − 2ν ∂ωz ; =− . ∂y ∂y 1−ν ∂x

3.7. If a material is incompressible (ν = 0.5), a state of hydrostatic stress σx x = σ yy = σzz produces no strain. One way to write the corresponding stress-strain relations is

Problems

49

σi j = 2μei j − qδi j , where q is an unknown hydrostatic pressure which will generally vary with position. Also, the condition of incompressibility requires that the dilatation e ≡ ekk = 0 . Show that under plain strain conditions, the stress components and the hydrostatic pressure q must satisfy the equations ∇ 2 q = div p and σx x + σ yy = −2q , where p is the body force.

Chapter 4

Stress Function Formulation

Newton’s law of gravitation states that two heavy bodies attract each other with a force proportional to the inverse square of their distance — thus it is essentially a vector theory, being concerned with forces. However, the idea of a scalar gravitational potential can be introduced by defining the work done in moving a unit mass from infinity to a given point in the field. The principle of conservation of energy requires that this be a unique function of position and it is easy to show that the gravitational force at any point is then proportional to the gradient of this scalar potential. Thus, the original vector problem is reduced to a problem about a scalar potential and its derivatives. In general, scalars are much easier to deal with than vectors. In particular, they lend themselves very easily to coördinate transformations, whereas vectors (and to an even greater extent tensors) require a set of special transformation rules (e.g. Mohr’s circle). In certain field theories, the scalar potential has an obvious physical significance. For example, in the conduction of heat, the temperature is a scalar potential in terms of which the vector heat flux can be defined. However, it is not necessary to the method that such a physical interpretation can be given. The gravitational potential can be given a physical interpretation as discussed above, but this interpretation may never feature in the solution of a particular problem, which is simply an exercise in the solution of a certain partial differential equation with appropriate boundary conditions. In the theory of elasticity, we make use of scalar potentials called stress functions or displacement functions which have no obvious physical meaning other than their use in defining stress or displacement components in terms of derivatives.

4.1 Choice of a Suitable Form In the choice of a suitable form for a stress or displacement function, there is only one absolute rule — that the operators which define the relationship between © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 J. R. Barber, Elasticity, Solid Mechanics and Its Applications 172, https://doi.org/10.1007/978-3-031-15214-6_4

51

52

4 Stress Function Formulation

the scalar and vector (or tensor) quantities should indeed define a vector (or tensor). For example, it is appropriate to define the displacement in terms of the first derivatives (the gradient) of a scalar or to define the stress components in terms of the second derivatives of a scalar, since the second derivatives of a scalar form the components of a Cartesian tensor. In effect, what we are doing in requiring this similarity of form between the definitions and the defined quantity is ensuring that the relationship is preserved in coördinate transformations. It would be quite possible to work out an elasticity problem in terms of the displacement components u x , u y , u z , treating these as essentially scalar quantities which vary with position — indeed this was a technique which was used in early theories. However, we would then get into trouble as soon as we tried to make any statements about quantities in other coördinate directions. By contrast, if we define (for example) u x = ∂ψ/∂x; u y = ∂ψ/∂ y; u z = ∂ψ/∂z, (i.e. u = ∇ψ), it immediately follows that u x = ∂ψ/∂x  for any x  .

4.2 The Airy Stress Function If there are no body forces and the non-zero stress components σx x , σx y , σ yy are independent of z, they must satisfy the two equilibrium equations ∂σx y ∂σx x + =0; ∂x ∂y

∂σ yx ∂σ yy + =0. ∂x ∂y

(4.1)

The two terms in each of these equations are therefore equal and opposite. Suppose we take the first term in the first equation and integrate it once with respect to x and once with respect to y to define a new function ψ(x, y). We can then write ∂σx y ∂σx x ∂2ψ =− ≡ , ∂x ∂y ∂x∂ y

(4.2)

and a particular solution is σx x =

∂ψ ; ∂y

σx y = −

∂ψ . ∂x

(4.3)

Substituting the second of these results in the second equilibrium equation and using the same technique to define another new function φ(x, y), we obtain ∂σ yy ∂σ yx ∂2ψ ∂3φ =− = , ≡ ∂y ∂x ∂x 2 ∂x 2 ∂ y so a particular solution is

(4.4)

4.2 The Airy Stress Function

53

σ yy =

∂2φ ∂φ . ; ψ= ∂x 2 ∂y

(4.5)

Using (4.3, 4.5) we can then express all three stress components in terms of φ as σx x =

∂2φ ∂2φ ∂2φ ; σ yy = ; σx y = − . 2 2 ∂y ∂x ∂x∂ y

(4.6)

This representation was introduced by G. B. Airy1 in 1862 and φ is therefore generally referred to as the Airy stress function. It is clear from the derivation that the stress components (4.6) will satisfy the two-dimensional equilibrium equations (4.1) for all functions φ. Also, the mathematical operations involved in obtaining (4.6) can be performed for any two-dimensional stress distribution satisfying (4.1) and hence the representation (4.6) is complete, meaning that it is capable of representing the most general such stress distribution. It is worth noting that the derivation makes no reference to Hooke’s law and hence the Airy stress function also provides a complete representation for two-dimensional problems involving inelastic constitutive laws — for example plasticity theory — where its satisfaction of the equilibrium equations remains an advantage. The technique introduced in this section can be used in other situations to define a representation of a field satisfying one or more partial differential equations. The trick is to rewrite the equation so that terms involving different unknowns are on opposite sides of the equals sign and then equate each side of the equation to a derivative that contains the lowest common denominator of the derivatives on the two sides, as in equations (4.2, 4.4). For other applications of this method, see Problems 4.3 and 4.4.

4.2.1 Transformation of coördinates It is easily verified that equation (4.6) transforms as a Cartesian tensor as required. For example, using (1.29) we can write   ∂ 2 ∂2φ ∂ + sin θ = cos θ φ ∂x ∂y ∂x  2 ∂2φ ∂2φ 2 ∂2φ 2 = sin θ cos θ , cos θ + sin θ + 2 ∂x 2 ∂ y2 ∂x∂ y from which using (1.11) and (4.6) we deduce that σ yy = ∂ 2 φ/∂x  2 as required.

1

For a good historical survey of the development of potential function methods in Elasticity, see H. M. Westergaard, Theory of Elasticity and Plasticity, Dover, New York, 1964, Chapter 2.

54

4 Stress Function Formulation

If we were simply to seek a representation of a two-dimensional Cartesian tensor in terms of the derivatives of a scalar function without regard to the equilibrium equations, the Airy stress function is not the most obvious form. It would seem more natural to write σx x = ∂ 2 ψ/∂x 2 ; σ yy = ∂ 2 ψ/∂ y 2 ; σx y = ∂ 2 ψ/∂x∂ y and indeed this also leads to a representation which is widely used as part of the general threedimensional solution (see Chapter 22 below). In some books, it is stated or implied that the Airy function is used because it satisfies the equilibrium equation, but this is rather misleading, since the ‘more obvious form’ — although it requires some constraints on ψ in order to satisfy equilibrium — can be shown to define displacements which automatically satisfy the compatibility condition which is surely equally useful2 . The real reason for preferring the Airy function is that it is complete, whilst the alternative is more restrictive, as we shall see in §21.1.

4.2.2 Non-zero body forces If the body force p is not zero, but is of a restricted form such that it can be written p = −∇V , where V is a two-dimensional scalar potential, equations (4.6) can be generalized by including an extra term +V in each of the normal stress components whilst leaving the shear stress definition unchanged. It is easily verified that, with this modification, the equilibrium equations are again satisfied. Problems involving body forces will be discussed in more detail in Chapter 7 below. For the moment, we restrict attention to the the case where p = 0 and hence where the representation (4.6) is appropriate.

4.3 The Governing Equation As discussed in Chapter 2, the stress or displacement field has to satisfy the equations of equilibrium and compatibility if they are to describe permissible states of an elastic body. In stress function representations, this generally imposes certain constraints on the choice of stress function, which can be expressed by requiring it to be a solution of a certain partial differential equation. We have already seen that the equilibrium condition is satisfied identically by the use of the Airy stress function φ, so it remains to determine the governing equation for φ by substituting the representation (4.6) into the compatibility equation in two dimensions.

2

See Problem 2.3.

4.3 The Governing Equation

55

4.3.1 The compatibility condition We first express the compatibility condition in terms of stresses using the stress-strain relations, obtaining ∂ 2 σ yy ∂ 2 σ yy ∂ 2 σx y ∂ 2 σx x ∂ 2 σx x + −ν − 2(1 + ν) −ν =0 , 2 2 2 ∂y ∂y ∂x∂ y ∂x ∂x 2

(4.7)

where we have cancelled a common factor of 1/E. We then substitute for the stress components from (4.6) obtaining ∂4φ ∂4φ ∂4φ ∂4φ ∂4φ − ν + 2(1 + ν) + − ν =0 , ∂ y4 ∂x 2 ∂ y 2 ∂x 2 ∂ y 2 ∂x 4 ∂x 2 ∂ y 2 i.e.

∂4φ ∂4φ ∂4φ + 2 + = ∂x 4 ∂x 2 ∂ y 2 ∂ y4



∂2 ∂2 + ∂x 2 ∂ y2

(4.8)

2 φ=0 .

This equation is known as the biharmonic equation and is usually written in the concise form ∇ 4φ = 0 . (4.9) The biharmonic equation is the governing equation for the Airy stress function in elasticity problems. Thus, by using the Airy stress function representation, the problem of determining the stresses in an elastic body is reduced to that of finding a solution of equation (4.9) (i.e. a biharmonic function) whose derivatives satisfy certain boundary conditions on the surfaces.

4.3.2 Method of solution Historically, the boundary-value problem for the Airy stress function has been approached in a semi-inverse way — i.e. by using the variation of tractions along the boundaries to give a clue to the kind of function required, but then exploring the stress fields developed from a wide range of such functions and selecting a combination which can be made to satisfy the required conditions. The disadvantage with this method is that it requires a wide experience of particular solutions and even then is not guaranteed to be successful. A more modern method which has the advantage of always developing an appropriate stress function if the boundary conditions are unmixed (i.e. all specified in terms of stresses or all in terms of displacements) is based on representing the stress function in terms of analytic functions of the complex variable. This method will be discussed in Chapter 20. However, although it is powerful and extremely elegant, it requires a certain familarity with the properties of functions of the complex variable and generally involves the evaluation of a contour integral, which may not be as

56

4 Stress Function Formulation

convenient a numerical technique as a direct series or finite difference attack on the original problem using real stress functions. There are some exceptions — notably for bodies which are susceptible of a simple conformal transformation into the unit circle.

4.3.3 Reduced dependence on elastic constants It is clear from dimensional considerations that, when the compatibility equation is expressed in terms of φ, Young’s modulus must appear in every term and can therefore be cancelled as in equation (4.7), but it is an unexpected bonus that the Poisson’s ratio terms also cancel in (4.8), leaving an equation that is independent of elastic constants. It follows that the stress field in a simply-connected elastic body3 in a state of plane strain or plane stress is independent of the material properties if the boundary conditions are expressed in terms of tractions and, in particular, that the plane stress and plane strain fields are identical. Dundurs4 has shown that a similar reduced dependence on elastic constants occurs in plane problems involving interfaces between two dissimilar elastic materials. In such cases, three independent dimensionless parameters (e.g. μ1 /μ2 , ν1 , ν2 ) can be formed from the elastic constants μ1 , ν1 , μ2 , ν2 of the materials 1, 2 respectively, but Dundurs proved that the stress field can be written in terms of only two parameters which he defined as     κ1 + 1 κ 2 + 1  κ1 + 1 κ 2 + 1 (4.10) − + α= μ1 μ2 μ1 μ2     κ1 − 1 κ2 − 1  κ1 + 1 κ2 + 1 , (4.11) β= − + μ1 μ2 μ1 μ2 where κ is Kolosov’s constant [see equation (3.11)]. It can be shown that Dundurs’ parameters must lie in the range −1 ≤ α ≤ 1, −0.5 ≤ β ≤ 0.5 if 0 ≤ ν1 , ν2 ≤ 0.5.

Problems 4.1. Newton’s law of gravitation states that two heavy particles of mass m 1 , m 2 respectively will experience a mutual attractive force

3

The restriction to simply-connected bodies is necessary, since in a multiply-connected body there is an implied displacement condition, as explained in §2.2 above. 4 J. Dundurs (1969), Discussion on ‘Edge bonded dissimilar orthogonal elastic wedges under normal and shear loading’, ASME Journal of Applied Mechanics, Vol. 36, pp. 650–652.

Problems

57

F=

γm 1 m 2 , R2

where R is the distance between the particles and γ is the gravitational constant. Use an energy argument and superposition to show that the force acting on a particle of mass m 0 can be written F = −γm 0 ∇V , 

where V (x, y, z) = −



ρ(ξ, η, ζ)dξdηdζ  , (x − ξ)2 + (y − η)2 + (z − ζ)2

 represents the volume of the universe and ρ is the density of material in the universe, which will generally be a function of position (ξ, η, ζ). Could a similar method have been used if Newton’s law had been of the more general form γm 1 m 2 . F= Rλ If so, what would have been the corresponding expression for V ? If not, why not? 4.2. An ionized liquid in an electric field experiences a body force p. Show that the liquid can be in equilibrium only if p is a conservative vector field. Hint: Remember that a stationary liquid must be everywhere in a state of hydrostatic stress. 4.3. An antiplane state of stress is one for which the only non-zero stress components are σzx , σzy and these are independent of z. Show that two of the three equilibrium equations are then satisfied identically if there is no body force. Use a technique similar to that of §4.2 to develop a representation of the non-zero stress components in terms of a scalar function, such that the remaining equilibrium equation is satisfied identically. 4.4. If a body of fairly general axisymmetric shape is loaded in torsion, the only non-zero stress components in cylindrical polar coördinates are σθr , σθz and these are required to satisfy the equilibrium equation 2σθr ∂σθz ∂σθr + + =0. ∂r r ∂z Use a technique similar to that of §4.2 to develop a representation of these stress components in terms of a scalar function, such that the equilibrium equation is satisfied identically. 4.5. (i) Show that the function φ = yω + ψ

58

4 Stress Function Formulation

satisfies the biharmonic equation if ω, ψ are both harmonic (i.e. ∇ 2 ω = 0 and ∇ 2 ψ = 0). (ii) Develop expressions for the stress components in terms of ω, ψ, based on the use of φ as an Airy stress function. (iii) Show that a solution suitable for the half-plane y > 0 subject to normal surface tractions only (i.e. σx y = 0 on y = 0) can be obtained by writing ω=−

∂ψ ∂y

and hence that under these conditions the normal stress σx x near the surface y = 0 is equal to the applied traction σ yy . (iv) Do you think this is a rigorous proof? Can you think of any exceptions? If so, at what point in your proof of section (iii) can you find a lack of generality? 4.6. The constitutive law for an orthotropic elastic material in plane stress can be written ex x = s11 σx x + s12 σ yy ; e yy = s12 σx x + s22 σ yy ; ex y = s44 σx y , where s11 , s12 , s22 , s44 are elastic constants. Using the Airy stress function φ to represent the stress components, find the equation that must be satisfied by φ. 4.7. Show that if the two-dimensional function ω(x, y) is harmonic (∇ 2 ω = 0), the function φ = (x 2 + y 2 ) ω will be biharmonic. 4.8. The constitutive law for an isotropic incompressible elastic material can be written σi j = σδi j + 2μ ei j , where σ=

σkk 3

represents an arbitrary hydrostatic stress field. Some soils can be approximated as an incompressible material whose modulus varies linearly with depth, so that μ = Mz for the half-space z > 0. Use the displacement function representation u = ∇φ

Problems

59

to develop a potential function solution for the stresses in such a body. Show that the functions φ, σ must satisfy the relations ∇ 2 φ = 0 ; σ = −2M

∂φ , ∂z

and hence obtain expressions for the stress components in terms of the single harmonic function φ. If the half-space is loaded by a normal pressure σzz (x, y, 0) = − p(x, y) ; σzx (x, y, 0) = σzy (x, y, 0) = 0 , show that the corresponding normal surface displacement u z (x, y, 0) is linearly proportional to the local pressure p(x, y) and find the constant of proportionality5 . 4.9. Show that Dundurs’ constant β → 0 for plane strain in the limit where ν1 = 0.5 and μ1 /μ2 → 0 — i.e. material 1 is incompressible and has a much lower shear modulus6 than material 2. What is the value of α in this limit? 4.10. Solve Problem 3.7 for the case where there is no body force, using the Airy stress function φ to represent the stress components. Hence show that the governing equation is ∇ 4 φ = 0, as in the case of compressible materials.

5

For an alternative proof of this result see C. R. Calladine and J. A. Greenwood (1978), Line and point loads on a non-homogeneous incompressible elastic half-space, Quarterly Journal of Mechanics and Applied Mathematics, Vol. 31, pp. 507–529. 6 This is a reasonable approximation for the important case of rubber (material 1) bonded to steel (material 2).

Chapter 5

Problems in Rectangular Coördinates

The Cartesian coördinate system (x, y) is clearly particularly suited to the problem of determining the stresses in a rectangular body whose boundaries are defined by equations of the form x = a, y = b. A wide range of such problems can be treated using stress functions which are polynomials in x, y. In particular, polynomial solutions can be obtained for ‘Mechanics of Materials’ type beam problems in which a rectangular bar is bent by an end load or by a distributed load on one or both faces.

5.1 Biharmonic Polynomial Functions In rectangular coördinates, the biharmonic equation takes the form ∂4φ ∂4φ ∂4φ +2 2 2 + =0 4 ∂x ∂x ∂ y ∂ y4

(5.1)

and it follows that any polynomial in x, y of degree less than four will be biharmonic and is therefore appropriate as a stress function. However, for higher order polynomial terms, equation (5.1) is not identically satisfied. Suppose, for example, that we consider just those terms in a general polynomial whose combined degree (the sum of the powers of x and y) is N . We can write these terms in the form PN (x, y) = A0 x N + A1 x N −1 y + A2 x N −2 y 2 + . . . + A N y N N  = Ai x N −i y i ,

(5.2)

i=0

Supplementary Information The online version contains supplementary material available at https://doi.org/10.1007/978-3-031-15214-6_5. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 J. R. Barber, Elasticity, Solid Mechanics and Its Applications 172, https://doi.org/10.1007/978-3-031-15214-6_5

61

62

5 Problems in Rectangular Coördinates

where we note that there are (N +1) independent coefficients, Ai (i = 0, N ). If we now substitute PN (x, y) into equation (5.1), we shall obtain a new polynomial of degree (N −4), since each term is differentiated four times. We can denote this new polynomial by Q N −4 (x, y) where Q N −4 (x, y) = ∇ 4 PN (x, y) N −4  Bi x (N −4−i) y i . =

(5.3) (5.4)

i=0

The (N −3) coefficients B0 , . . . , B N −4 are easily obtained by expanding the righthand side of equation (5.3) and equating coefficients. For example, B0 = N (N − 1)(N − 2)(N − 3)A0 + 4(N − 2)(N − 3)A2 + 24 A4 .

(5.5)

Now the original function PN (x, y) will be biharmonic if and only if Q N −4 (x, y) is zero for all x, y and this in turn is only possible if every term in the series (5.4) is identically zero, since the polynomial terms are all linearly independent of each other. In other words Bi = 0 ; i = 0 to (N − 4) . These conditions can be converted into a corresponding set of (N −3) equations for the coefficients Ai . For example, the equation B0 = 0 gives N (N − 1)(N − 2)(N − 3)A0 + 4(N − 2)(N − 3)A2 + 24 A4 = 0 ,

(5.6)

from (5.5). We shall refer to the (N −3) equations of this form as constraints on the coefficients Ai , since the coefficients are constrained to satisfy them if the original polynomial is to be biharmonic. One approach would be to use the constraint equations to eliminate (N −3) of the unknown coefficients in the original polynomial — for example, we could treat the first four coefficients, A0 , A1 , A2 , A3 , as unknown constants and use the constraint equations to define all the remaining coefficients in terms of these unknowns. Equation (5.6) would then be treated as an equation for A4 and the subsequent constraint equations would each define one new constant in the series. It may help to consider a particular example at this stage. Suppose we consider the fifth degree polynomial P5 (x, y) = A0 x 5 + A1 x 4 y + A2 x 3 y 2 + A3 x 2 y 3 + A4 x y 4 + A5 y 5 , which has six independent coefficients. Substituting into equation (5.3), we obtain the first degree polynomial Q 1 (x, y) = (120 A0 + 24 A2 + 24 A4 )x + (24 A1 + 24 A3 + 120 A5 )y . The coefficients of x and y in Q 1 must both be zero if P5 is to be biharmonic and we can write the resulting two constraint equations in the form

5.1 Biharmonic Polynomial Functions

A4 = −5A0 − A2 ;

63

A5 = −

A3 A1 − . 5 5

(5.7)

Finally, we use (5.7) to eliminate A4 , A5 in the original definition of P5 , obtaining the definition of the most general biharmonic fifth degree polynomial     y5 P5 (x, y) = A0 x 5 − 5x y 4 + A1 x 4 y − 5    3 2  y5 4 2 3 +A2 x y − x y + A3 x y − . 5

(5.8)

This function will be biharmonic for any values of the four independent constants A0 , A1 , A2 , A3 . We can express this by stating that the biharmonic polynomial P5 has four degrees of freedom. In general, the polynomial Q is of degree 4 less than P because the biharmonic equation is of degree 4. It follows that there are always four fewer constraint equations than there are coefficients in the original polynomial P and hence that they can be satisfied leaving a polynomial with 4 degrees of freedom. However, the process degenerates if N < 3. In view of the above discussion, it might seem appropriate to write an expression for the general polynomial of degree N in the form of equation (5.8) as a preliminary to the solution of polynomial problems in rectangular coördinates. However, as can be seen from equation (5.8), the resulting expressions are algebraically messy and this approach becomes unmanageable for problems of any complexity. Instead, it turns out to be more straightforward algebraically to define problems in terms of the simpler unconstrained polynomials like equation (5.2) and to impose the constraint equations at a later stage in the solution.

5.1.1 Second and third degree polynomials We recall that the stress components are defined in terms of the stress function φ through the relations σx x =

∂2φ ∂2φ ∂2φ . ; σ = ; σ = − yy x y ∂ y2 ∂x 2 ∂x∂ y

(5.9)

It follows that when the stress function is a polynomial of degree N in x, y, the stress components will be polynomials of degree (N −2). In particular, constant and linear terms in φ correspond to null stress fields (zero stress everywhere) and can be disregarded. The second degree polynomial φ = A0 x 2 + A1 x y + A2 y 2 yields the stress components

64

5 Problems in Rectangular Coördinates

σx x = 2 A2 ; σx y = −A1 ; σ yy = 2 A0 and hence corresponds to the most general state of biaxial uniform stress. The third degree polynomial φ = A0 x 3 + A1 x 2 y + A2 x y 2 + A3 y 3

(5.10)

yields the stress components σx x = 2 A2 x + 6A3 y ; σx y = −2 A1 x − 2 A2 y ; σ yy = 6A0 x + 2 A1 y . If we arbitrarily set A0 , A1 , A2 = 0, the only remaining non-zero stress will be σx x = 6A3 y , which corresponds to a state of pure bending, when applied to the rectangular beam −a < x < a, −b < y < b, as shown in Figure 5.1.

Fig. 5.1 The rectangular beam in pure bending.

The other terms in equation (5.10) correspond to a more general state of bending. For example, the constant A0 describes bending of the beam by tractions σ yy applied to the boundaries y = ±b, whilst the terms involving shear stresses σx y could be obtained by describing a general state of biaxial bending with reference to a Cartesian coördinate system which is not aligned with the axes of the beam. The above solutions are of course very elementary, but we should remember that, in contrast to the Mechanics of Materials solutions for simple bending, they are obtained without making any simplifying assumptions about the stress fields. For example, we have not assumed that plane sections remain plane, nor have we demanded that the beam be long in comparison with its depth. Thus, the present section could be taken as verifying the exactness of the Mechanics of Materials solutions for uniform stress and simple bending, as applied to a rectangular beam.

5.2 Rectangular Beam Problems

65

5.2 Rectangular Beam Problems 5.2.1 Bending of a beam by an end load Figure 5.2 shows a rectangular beam, 0 < x < a, −b < y < b, subjected to a transverse force, F at the end x = 0, and built-in at the end x = a, the horizontal boundaries y = ±b being traction free. The boundary conditions for this problem are most naturally written in the form



b

−b

σx y = 0 ; y = ±b σ yy = 0 ; y = ±b

(5.11) (5.12)

σx x = 0 ; x = 0

(5.13)

σx y dy = F ; x = 0 .

(5.14)

Fig. 5.2 Cantilever with an end load.

The boundary condition (5.14) is imposed in the weak form, which means that the value of the traction is not specified at each point on the boundary — only the force resultant is specified. In general, we shall find that problems for the rectangular beam have finite polynomial solutions when the boundary conditions on the ends are stated in the weak form, but that the strong (i.e. pointwise) boundary condition can only be satisfied on all the boundaries by an infinite series or transform solution. This issue is further discussed in Chapter 6. Mechanics of Materials considerations suggest that the bending moment in this problem will vary linearly with x and hence that the stress component σx x will have a leading term proportional to x y. This in turn suggests a fourth degree polynomial term x y 3 in the stress function φ. Our procedure is therefore to start with the trial stress function (5.15) φ = C1 x y 3 ,

66

5 Problems in Rectangular Coördinates

examine the corresponding tractions on the boundaries and then seek a corrective solution which, when superposed on equation (5.15), yields the solution to the problem. Substituting (5.15) into (5.9), we obtain σx x = 6C1 x y σx y = −3C1 y 2 σ yy = 0 ,

(5.16) (5.17) (5.18)

from which we note that the boundary conditions (5.12, 5.13) are satisfied identically, but that (5.11) is not satisfied, since (5.17) implies the existence of an unwanted uniform shear traction −3C1 b2 on both of the edges y = ±b. This unwanted traction can be removed by superposing an appropriate uniform shear stress, through the additional stress function term C2 x y. Thus, if we define φ = C1 x y 3 + C2 x y ,

(5.19)

equations (5.16, 5.18) remain unchanged, whilst (5.17) is modified to σx y = −3C1 y 2 − C2 . The boundary condition (5.11) can now be satisfied if we choose C2 to satisfy the equation C2 = −3C1 b2 , so that σx y = 3C1 (b2 − y 2 ) .

(5.20)

The constant C1 can then be determined by substituting (5.20) into the remaining boundary condition (5.14), with the result C1 =

F . 4b3

The final stress field is therefore defined through the stress function φ=

F(x y 3 − 3b2 x y) , 4b3

(5.21)

the corresponding stress components being σx x =

3F x y 3F(b2 − y 2 ) ; σ = ; σ yy = 0 . x y 2b3 4b3

(5.22)

The solution of this problem is given in the Mathematica and Maple files ‘S521’.

5.2 Rectangular Beam Problems

67

Boundary conditions at x = a We note that no boundary conditions have been specified on the built-in end, x = a. In the weak form, these would be 

b

σx x dy = 0 ; x = a

(5.23)

σx y dy = F ; x = a

(5.24)

σx x ydy = Fa ; x = a .

(5.25)

−b b

 

−b b

−b

However, if conditions (5.11–5.14) are satisfied, (5.23–5.25) are merely equivalent to the condition that the whole beam be in equilibrium. Now the Airy stress function is so defined that whatever stress function is used, the corresponding stress field will satisfy equilibrium in the local sense of equations (2.4). Furthermore, if every particle of a body is separately in equilibrium, it follows that the whole body will also be in equilibrium. It is therefore not necessary to enforce equations (5.23–5.25), since if we were to check them, we should necessarily find that they are satisfied identically.

5.2.2 Higher order polynomials — a general strategy In the previous section, we developed the solution by trial and error, starting from the leading term whose form was dictated by equilibrium considerations. A more general technique is to identify the highest order polynomial term from equilibrium considerations and then write down the most general polynomial of that degree and below. The constant multipliers on the various terms are then obtained by imposing boundary conditions and biharmonic constraint equations. The only objection to this procedure is that it involves a lot of algebra. For example, in the problem of §5.2.1, we would have to write down the most general polynomial of degree 4 and below, which involves 12 separate terms even when we exclude the linear and constant terms as being null. However, this is not a serious difficulty if we are using Maple or Mathematica, so we shall first develop the steps needed for this general strategy. Shortcuts which would reduce the complexity of the algebra in a manual calculation will then be discussed in §5.2.3.

68

5 Problems in Rectangular Coördinates

Order of the polynomial Suppose we have a normal traction on the surface y = b varying with x n . In Mechanics of Materials terms, this corresponds to a distributed load proportional to x n and elementary equilibrium considerations show that the bending moment can then be expected to contain a term proportional to x n+2 . This in turn implies a bending stress σx x proportional to x n+2 y and a term in the stress function proportional to x n+2 y 3 — i.e. a term of polynomial order (n+5). A corresponding argument for shear tractions proportional to x m shows that we require a polynomial order of (m +4). We shall show in Chapter 30 that these arguments from equilibrium and elementary bending theory define the highest order of polynomial required to satisfy any polynomial boundary conditions on the lateral surfaces of a beam, even in threedimensional problems. A sufficient polynomial order can therefore be selected by the following procedure:(i) Identify the highest order polynomial term n in the normal tractions σ yy on the surfaces y = ±b. (ii) Identify the highest order polynomial term m in the shear tractions σ yx on the surfaces y = ±b. (iii) Use a polynomial for φ including all polynomial terms of order max(m + 4, n + 5) and below, but excluding constant and linear terms. In the special case where both surfaces y = ±b are traction free, it is sufficient to use a polynomial of 4th degree and below (as in §5.2.1).

Solution procedure Once an appropriate polynomial has been identified for φ, we proceed as follows:(i) Substitute φ into the biharmonic equation (5.1), leading to a set of constraint equations, as in §5.1. (ii) Substitute φ into equations (5.9), to obtain the stress components as functions of x, y. (iii) Substitute the equations defining the boundaries (e.g. x = 0, y = b, y = −b in the problem of §5.2.1) into appropriate1 stress components, to obtain the tractions on each boundary. (iv) For the longer boundaries (where strong boundary conditions will be enforced), sort the resulting expressions into powers of x or y and equate coefficients with the corresponding expression for the prescribed tractions.

Recall from §1.1.1 that the only stress components that act on (e.g.) y = b are those which contain y as one of the suffices.

1

5.2 Rectangular Beam Problems

69

(v) For the shorter boundaries, substitute the tractions into the appropriate weak boundary conditions, obtaining three further independent algebraic equations. The equations so obtained will generally not all be linearly independent, but they will be sufficient to determine all the coefficients uniquely. The solvers in Maple and Mathematica can handle this redundancy.

Example We illustrate this procedure with the example of Figure 5.3, in which a rectangular beam −a < x < a, −b < y < b is loaded by a uniform compressive normal traction p on y = b and simply supported at the ends.

Fig. 5.3 Simply supported beam with a uniform load.

The boundary conditions on the surfaces y = ±b can be written σ yx = 0 ; y = ±b

(5.26)

σ yy = − p ; y = b σ yy = 0 ; y = −b .

(5.27) (5.28)

These boundary conditions are to be satisfied in the strong sense. To complete the problem definition, we shall require three linearly independent weak boundary conditions on one or both of the ends x = ±a. We might use symmetry and equilibrium to argue that the load will be equally shared between the supports, leading to the conditions2 2

It is not necessary to use symmetry arguments to obtain three linearly independent weak conditions. Since the beam is simply supported, we know that  M(a) =  Fx (a) =

b −b b −b

 σx x (a, y)ydy = 0 ; M(−a) = σx x (a, y)dy = 0 .

b −b

σx x (−a, y)ydy = 0

70

5 Problems in Rectangular Coördinates

 Fx (a) =

b −b b

σx x (a, y)dy = 0

(5.29)

σx y (a, y)dy = pa

(5.30)

 Fy (a) =  M(a) =

−b b −b

σx x (a, y)ydy = 0

(5.31)

on the end x = a. As explained in §5.2.1, we do not need to enforce the additional three weak conditions on x = −a. The normal traction is uniform — i.e. it varies with x 0 (n = 0), so the above criterion demands a polynomial of order (n + 5) = 5. We therefore write φ = C1 x 2 + C2 x y + C3 y 2 + C4 x 3 + C5 x 2 y + C6 x y 2 + C7 y 3 + C8 x 4 +C9 x 3 y + C10 x 2 y 2 + C11 x y 3 + C12 y 4 + C13 x 5 + C14 x 4 y +C15 x 3 y 2 + C16 x 2 y 3 + C17 x y 4 + C18 y 5 .

(5.32)

This is a long expression, but remember we only have to type it in once to the computer file. We can cut and paste the expression in the solution of subsequent problems (and the reader can indeed cut and paste from the web file ‘polynomial’). Substituting (5.32) into the biharmonic equation (5.1), we obtain (120C13 + 24C15 + 24C17 )x + (24C14 + 24C16 + 120C18 )y +(24C8 + 8C10 + 24C12 ) = 0 and this must be zero for all x, y leading to the three constraint equations 120C13 + 24C15 + 24C17 = 0

(5.33)

24C14 + 24C16 + 120C18 = 0 24C8 + 8C10 + 24C12 = 0 .

(5.34) (5.35)

The stresses are obtained by substituting (5.32) into (5.9) with the result σx x = 2C3 + 2C6 x + 6C7 y + 2C10 x 2 + 6C11 x y + 12C12 y 2 + 2C15 x 3 σx y

+6C16 x 2 y + 12C17 x y 2 + 20C18 y 3 = −C2 − 2C5 x − 2C6 y − 3C9 x 2 − 4C10 x y − 3C11 y 2 − 4C14 x 3

σ yy

−6C15 x 2 y − 6C16 x y 2 − 4C17 y 3 = 2C1 + 6C4 x + 2C5 y + 12C8 x 2 + 6C9 x y + 2C10 y 2 + 20C13 x 3 +12C14 x 2 y + 6C15 x y 2 + 2C16 y 3 .

The tractions on y = b are therefore It is easy to verify that these conditions lead to the same solution as (5.29–5.31).

5.2 Rectangular Beam Problems

71

σ yx = −4C14 x 3 − (3C9 + 6C15 b)x 2 − (2C5 + 4C10 b + 6C16 b2 )x σ yy

−(C2 + 2C6 b + 3C11 b2 + 4C17 b3 ) = 20C13 x 3 + (12C8 + 12C14 b)x 2 + (6C4 + 6C9 b + 6C15 b2 )x +(2C1 + 2C5 b + +2C10 b2 + 2C16 b3 )

and these must satisfy equations (5.26, 5.27) for all x, giving 4C14 = 0 3C9 + 6C15 b = 0

(5.36) (5.37)

2C5 + 4C10 b + 6C16 b2 = 0 C2 + 2C6 b + 3C11 b2 + 4C17 b3 = 0

(5.38) (5.39)

20C13 = 0 12C8 + 12C14 b = 0 6C4 + 6C9 b + 6C15 b2 = 0

(5.40) (5.41) (5.42)

2C1 + 2C5 b + 2C10 b2 + 2C16 b3 = − p .

(5.43)

A similar procedure for the edge y = −b yields the additional equations 3C9 − 6C15 b = 0

(5.44)

2C5 − 4C10 b + 6C16 b = 0 C2 − 2C6 b + 3C11 b2 − 4C17 b3 = 0

(5.45) (5.46)

12C8 − 12C14 b = 0 6C4 − 6C9 b + 6C15 b2 = 0 2C1 − 2C5 b + 2C10 b2 − 2C16 b3 = 0 .

(5.47) (5.48) (5.49)

2

On x = a, we have σx x = 2C3 + 2C6 a + 6C7 y + 2C10 a 2 + 6C11 ay + 12C12 y 2 + 2C15 a 3 +6C16 a 2 y + 12C17 ay 2 + 20C18 y 3 σx y = −C2 − 2C5 a − 2C6 y − 3C9 a 2 − 4C10 ay − 3C11 y 2 − 4C14 a 3 −6C15 a 2 y − 6C16 ay 2 − 4C17 y 3 . Substituting into the weak conditions (5.29–5.31) and evaluating the integrals, we obtain the three additional equations 4C3 b + 4C6 ab + 4C10 a 2 b + 8C12 b3 + 4C15 a 3 b + 8C17 ab3 = 0 (5.50) −2C2 b − 4C5 ab − 6C9 a 2 b − 2C11 b3 − 8C14 a 3 b − 4C16 ab3 = pa (5.51) 4C7 b3 + 4C11 ab3 + 4C16 a 2 b3 + 8C18 b5 = 0 . (5.52) Finally, we solve equations (5.33–5.35, 5.36–5.49, 5.50–5.52) for the unknown constants C1 , ..., C18 and substitute back into (5.32), obtaining

72

5 Problems in Rectangular Coördinates

φ=

p  2 3 5 2 2 2 3 2 3 3 2 5x . y − y − 15b x y − 5a y + 2b y − 10b x 40b3

(5.53)

The corresponding stress field is p  2 3 2 2 15x y − 10y − 15a y + 6b y 20b3 3 px  2 2 b = − y 4b3  p = 3 y 3 − 3b2 y − 2b3 . 4b

σx x =

(5.54)

σx y

(5.55)

σ yy

(5.56)

The reader is encouraged to run the Maple or Mathematica files ‘S522’, which contain the above solution procedure. Notice that most of the algebraic operations are generated by quite simple and repetitive commands. These will be essentially similar for any polynomial problem involving rectangular coördinates, so it is a simple matter to modify the program to cover other cases.

5.2.3 Manual solutions — symmetry considerations If the solution is to be obtained manually, the complexity of the algebra makes the process time consuming and increases the likelihood of errors. Fortunately, the complexity can be reduced by utilizing the natural symmetry of the rectangular beam. In many problems, the loading has some symmetry which can be exploited in limiting the number of independent polynomial terms and even when this is not the case, some saving of complexity can be achieved by representing the loading as the sum of symmetric and antisymmetric parts. We shall illustrate this procedure by repeating the solution of the problem of Figure 5.3. The problem is symmetrical about the mid-point of the beam and hence, taking the origin there, we deduce that the resulting stress function will contain only even powers of x. This immediately reduces the number of terms in the general stress function to 10. The beam is also symmetrical about the axis y = 0, but the loading is not. We therefore decompose the problem into the two sub-problems illustrated in Figure 5.4(a,b). The problem in Figure 5.4(a) is antisymmetric in y and hence requires a stress function with only odd powers of y, whereas that of Figure 5.4(b) is symmetric and requires only even powers. In fact, the problem of Figure 5.4(b) clearly has the trivial solution corresponding to uniform uniaxial compression, σ yy = − p/2, the appropriate stress function being φ = − px 2 /4. For the problem of Figure 5.4(a), the most general fifth degree polynomial which is even in x and odd in y can be written

5.2 Rectangular Beam Problems

73

Fig. 5.4 Decomposition of the problem into (a) antisymmetric and (b) symmetric parts.

φ = C5 x 2 y + C7 y 3 + C14 x 4 y + C16 x 2 y 3 + C18 y 5 ,

(5.57)

which has just five degrees of freedom. We have used the same notation for the remaining constants as in (5.32) to aid in comparing the two solutions. The appropriate boundary conditions for this sub-problem are

 

σx y = 0 ; y = ±b p σ yy = ∓ ; y = ±b 2 b

(5.59)

σx x dy = 0 ; x = ±a

(5.60)

σx x ydy = 0 ; x = ±a .

(5.61)

−b b

−b

(5.58)

Notice that, in view of the symmetry, it is only necessary to satisfy these conditions on one of each pair of edges (e.g. on y = b, x = a). For the same reason, we do not have to impose a condition on the vertical force at x = ±a, since the symmetry demands that the forces be equal at the two ends and the total force must be 2 pa to preserve global equilibrium, this being guaranteed by the use of the Airy stress function, as in the problem of §5.2.1. It is usually better strategy to start a manual solution with the strong boundary conditions (equations [5.58, 5.59)], and in particular with those conditions that are homogeneous [in this case equation (5.58)], since these will often require that one or more of the constants be zero, reducing the complexity of subsequent steps. Substituting (5.57) into (5.9)2,3 , we find σx y = −2C5 x − 4C14 x 3 − 6C16 x y 2

(5.62)

σ yy = 2C5 y + 12C14 x y + 2C16 y .

(5.63)

2

3

74

5 Problems in Rectangular Coördinates

Thus, condition (5.58) requires that 4C14 x 3 + (2C5 + 6C16 b2 )x = 0 ; for all x and this condition is satisfied if and only if C14 = 0 and 2C5 + 6C16 b2 = 0 .

(5.64)

A similar procedure with equation (5.63) and boundary condition (5.59) gives the additional equation p (5.65) 2C5 b + 2C16 b3 = − . 2 Equations (5.64, 5.65) have the solution C5 = −

3p p ; C16 = 3 . 8b 8b

(5.66)

We next determine C18 from the condition that the function φ is biharmonic, obtaining (5.67) (24C14 + 24C16 + 120C18 )y = 0 and hence C18 = −

p , 40b3

(5.68)

from (5.64, 5.66, 5.67). It remains to satisfy the two weak boundary conditions (5.60, 5.61) on the ends x = ±a. The first of these is satisfied identically in view of the antisymmetry of the stress field and the second gives the equation 4C7 b3 + 4C16 a 2 b3 + 8C18 b5 = 0 , which, with equations (5.66, 5.68), serves to determine the remaining constant, C7 =

p(2b2 − 5a 2 ) . 40b3

The final solution of the complete problem (the sum of that for Figures 5.4(a) and (b)) is therefore obtained from the stress function φ=

p (5x 2 y 3 − y 5 − 15b2 x 2 y − 5a 2 y 3 + 2b2 y 3 − 10b3 x 2 ) , 40b3

as in the ‘computer solution’ (5.53), and the stresses are therefore given by (5.54– 5.56) as before.

5.3 Fourier Series and Transform Solutions

75

5.3 Fourier Series and Transform Solutions Polynomial solutions can, in principle, be extended to more general loading of the beam edges, as long as the tractions are capable of a power series expansion. However, the practical use of this method is limited by the algebraic complexity encountered for higher order polynomials and by the fact that many important traction distributions do not have convergent power series representations. A more useful method in such cases is to build up a general solution by components of Fourier form. For example, if we write φ = f (y) cos(λx) or φ = f (y) sin(λx) ,

(5.69)

substitution in the biharmonic equation (5.1) shows that f (y) must have the general form (5.70) f (y) = (A + By)eλy + (C + Dy)e−λy , where A, B, C, D are arbitrary constants. Alternatively, by defining new arbitrary constants A , B  , C  , D  through the relations A = (A +C  )/2, B = (B  + D  )/2, C = (A −C  )/2, D = (B  − D  )/2, we can group the exponentials into hyperbolic functions, obtaining the equivalent form f (y) = (A + B  y) cosh(λy) + (C  + D  y) sinh(λy) .

(5.71)

The hyperbolic form enables us to take advantage of any symmetry about the line y = 0, since cosh(λy), y sinh(λy) are even functions of y and sinh(λy), y cosh(λy) are odd functions. More general biharmonic stress functions can be constructed by superposition of terms like (5.70, 5.71), leading to Fourier series expansions for the tractions on the surfaces y = ±b. The theory of Fourier series can then be used to determine the coefficients in the series, using strong boundary conditions on y = ±b. Quite general traction distributions can be expanded in this way, so Fourier series solutions provide a methodology applicable to any problem for the rectangular bar.

5.3.1 Choice of form The stresses due to the stress function φ = f (y) cos(λx) are σx x = f  (y) cos(λx) ; σx y = λ f  (y) sin(λx) ; σ yy = −λ2 f (y) cos(λx) and the tractions on the edge x = a are σx x (a, y) = f  (y) cos(λa) ; σx y (a, y) = λ f  (y) sin(λa) .

(5.72)

76

5 Problems in Rectangular Coördinates

It follows that we can satisfy homogeneous boundary conditions on one (but not both) of these tractions in the strong sense, by restricting the Fourier series to specific values of λ. In equation (5.72), the choice λ = nπ/a will give σx y = 0 on x = ±a, whilst λ = (2n − 1)π/2a will give σx x = 0 on x = ±a, where n is any integer.

Example We illustrate this technique by considering the rectangular beam −a < x < a, −b < y < b, simply supported at x = ±a and loaded by compressive normal tractions p1 (x) on the upper edge y = b and p2 (x) on y = −b — i.e. σx y = 0

; y = ±b

(5.73)

σ yy = − p1 (x) ; y = b = − p2 (x) ; y = −b

(5.74) (5.75)

σx x = 0

(5.76)

; x = ±a .

Notice that we have replaced the weak conditions (5.60, 5.61) by the strong condition (5.76). As in §5.2.2, it is not necessary to enforce the remaining weak conditions (those involving the vertical forces on x = ±a), since these will be identically satisfied by virtue of the equilibrium condition. The algebraic complexity of the problem will be reduced if we use the geometric symmetry of the beam to decompose the problem into four sub-problems. For this purpose, we define 1

p1 (x) + 4 1

f 2 (x) = − f 2 (−x) ≡ p1 (x) − 4

1 f 3 (x) = f 3 (−x) ≡ p1 (x) + 4

1 f 4 (x) = − f 4 (−x) ≡ p1 (x) − 4 f 1 (x) = f 1 (−x) ≡

p1 (−x) + p2 (x) + p2 (−x) p1 (−x) + p2 (x) − p2 (−x) p1 (−x) − p2 (x) − p2 (−x) p1 (−x) − p2 (x) + p2 (−x)

 (5.77)  (5.78)  (5.79)  (5.80)

and hence p1 (x)= f 1 (x)+ f 2 (x) + f 3 (x) + f 4 (x) ; p2 (x) = f 1 (x)+ f 2 (x)− f 3 (x)− f 4 (x) . The boundary conditions now take the form σx y = 0 ; y = ±b σ yy = − f 1 (x) − f 2 (x) − f 3 (x) − f 4 (x) ; y = b σx x

= − f 1 (x) − f 2 (x) + f 3 (x) + f 4 (x) ; y = −b = 0 ; x = ±a

5.3 Fourier Series and Transform Solutions

77

and each of the functions f 1 , f 2 , f 3 , f 4 defines a separate problem with either symmetry or antisymmetry about the x- and y-axes. We shall here restrict attention to the loading defined by the function f 3 (x), which is symmetric in x and antisymmetric in y. The boundary conditions of this sub-problem are ; y = ±b σx y = 0 σ yy = ∓ f 3 (x) ; y = ±b

(5.81) (5.82)

σx x = 0

(5.83)

; x = ±a .

The problem of equations (5.81–5.83) is even in x and odd in y, so we use a cosine series in x with only the odd terms from the hyperbolic form (5.71) — i.e. φ=

∞



 An y cosh(λn y) + Bn sinh(λn y) cos(λn x) ,

(5.84)

n=1

where An , Bn are arbitrary constants. The strong condition (5.83) on x = ±a can then be satisfied in every term by choosing λn =

(2n − 1)π . 2a

(5.85)

The corresponding stresses are σx x =

∞

  2 An λn sinh(λn y)+ An λ2n y cosh(λn y)+ Bn λ2n sinh(λn y) cos(λn x) n=1

σx y =

∞



 An λn cosh(λn y)+ An λ2n y sinh(λn y)+ Bn λ2n cosh(λn y) sin(λn x)

n=1

σ yy = −

∞



 An λ2n y cosh(λn y) + Bn λ2n sinh(λn y) cos(λn x)

(5.86)

n=1

and hence the boundary conditions (5.81, 5.82) on y = ±b require that ∞



 An λn cosh(λn b)+ An λ2n b sinh(λn b)+ Bn λ2n cosh(λn b) sin(λn x) = 0

n=1

(5.87) ∞



 An λ2n b cosh(λn b)+ Bn λ2n sinh(λn b) cos(λn x) = f 3 (x) .

(5.88)

n=1

To invert the series, we multiply (5.88) by cos(λm x) and integrate from −a to a, obtaining

78

5 Problems in Rectangular Coördinates ∞   n=1

a

−a



 An λ2n b cosh(λn b) + Bn λ2n sinh(λn b) cos(λn x) cos(λm x)d x  =

a

−a

f 3 (x) cos(λm x)d x .

(5.89)

The integrals on the left-hand side are all zero except for the case m = n and hence, evaluating the integrals, we find  a

 2 2 Am λm b cosh(λm b) + Bm λm sinh(λm b) a = f 3 (x) cos(λm x)d x . (5.90) −a

The homogeneous equation (5.87) is clearly satisfied if Am λm cosh(λm b) + Am λ2m b sinh(λm b) + Bm λ2m cosh(λm b) = 0 .

(5.91)

Solving (5.90, 5.91) for Am , Bm , we have

 a cosh(λm b) Am = f 3 (x) cos(λm x)d x λm a[λm b − sinh(λm b) cosh(λm b)] −a  a (cosh(λm b) + λm b sinh(λm b)) Bm = − 2 f 3 (x) cos(λm x)d x , λm a[λm b − sinh(λm b) cosh(λm b)] −a (5.92)

where λm is given by (5.85). The stresses are then recovered by substitution into equations (5.86). The corresponding solutions for the functions f 1 , f 2 , f 4 are obtained in a similar way, but using a sine series for the odd functions f 2 , f 4 and the even terms y sinh(λy), cosh(λy) in φ for f 1 , f 2 . The complete solution is then obtained by superposing the solutions of the four sub-problems. The Fourier series method is particularly useful in problems where the traction distributions on the long edges have no power series expansion, typically because of discontinuities in the loading. For example, suppose the beam is loaded only by a concentrated compressive force F on the upper edge at x = 0, corresponding to the loading p1 (x) = Fδ(x), p2 (x) = 0 in equations (5.78, 5.79). For the symmetric/antisymmetric sub-problem considered above, we then have f 3 (x) =

Fδ(x) 2

from (5.79), and the integral in equations (5.92) is therefore  a F f 3 (x) cos(λm x)d x = , 2 −a for all m.

5.3 Fourier Series and Transform Solutions

79

This solution satisfies the end condition on σx x in the strong sense, but the condition on σx y only in the weak sense. In other words, the tractions σx y on the ends add up to the forces required to maintain equilibrium, but we have no control over the exact distribution of these tractions. This represents an improvement over the polynomial solution of §5.2.3, where weak conditions were used for both end tractions, so we might be tempted to use a Fourier series even for problems with continuous polynomial loading. However, this improvement is made at the cost of an infinite series solution. If the series were truncated at a finite value of n, errors would be obtained particularly near the ends or any discontinuities in the loading.

5.3.2 Fourier transforms If the beam is infinite or semi-infinite (a → ∞), the series (5.84) must be replaced by the integral representation 

∞

φ(x, y) =

f (λ, y) cos(λx)dλ ,

(5.93)

0

where f (λ, y) = A(λ)y cosh(λy) + B(λ) sinh(λy) . Equation (5.93) is introduced here as a generalization of (5.69) by superposition, but φ(x, y) is in fact the Fourier cosine transform of f (λ, y), the corresponding inversion being  2 ∞ φ(x, y) cos(λx)d x . f (λ, y) = π 0 The boundary conditions on y = ±b will also lead to Fourier integrals, which can be inverted in the same way to determine the functions A(λ), B(λ). For a definitive treatment of the Fourier transform method, the reader is referred to the treatise by Sneddon3 . Extensive tables of Fourier transforms and their inversions are given by Erdelyi4 . The cosine transform (5.93) will lead to a symmetric solution. For more general loading, the complex exponential transform can be used. It is worth remarking on the way in which the series and transform solutions are natural generalizations of the elementary solution (5.69). One of the most powerful techniques in Elasticity — and indeed in any physical theory characterized by linear partial differential equations — is to seek a simple form of solution (often in

3

I. N. Sneddon, Fourier Transforms, McGraw-Hill, New York, 1951. A. Erdelyi, ed., Tables of Integral Transforms, Bateman Manuscript Project, California Institute of Technology, Vol. 1, McGraw-Hill, New York, 1954. 4

80

5 Problems in Rectangular Coördinates

separated-variable form) containing a parameter which can take a range of values. A more general solution can then be developed by superposing arbitrary multiples of the solution with different values of the parameter. For example, if a particular solution can be written symbolically as φ = f (x, y, λ), where λ is a parameter, we can develop a general series form φ(x, y) =

∞ 

Ai f (x, y, λi ) ,

(5.94)

i=0

or an integral form

 φ(x, y) =

b

A(λ) f (x, y, λ)dλ .

a

The series form will naturally arise if there is a discrete set of eigenvalues, λi for which f (x, y, λi ) satisfies some of the boundary conditions of the problem. Additional examples of this kind will be found in §§6.2, 11.2. In this case, the series (5.94) is most properly seen as an eigenfunction expansion. Integral forms arise most commonly (but not exclusively) in problems involving infinite or semi-infinite domains (see, for example, §§11.3, 32.2.2). Any particular solution containing a parameter can be used in this way and, since transforms are commonly named after their originators, the reader desirous of instant immortality might like to explore some of those which have not so far been used. Of course, the usefulness of the resulting solution depends upon its completeness — i.e. its capacity to represent all stress fields of a given class — and upon the ease with which the transform can be inverted.

Problems 5.1. The beam −b < y < b, 0 < x < L, is built-in at the end x = 0 and loaded by a uniform shear traction σx y = S on the upper edge, y = b, the remaining edges, x = L , y = −b being traction free. Find a suitable stress function and the corresponding stress components for this problem, using the weak boundary conditions on x = L. 5.2. The beam −b < y < b, −L < x < L is simply supported at the ends x = ±L and loaded by a shear traction σx y = Sx/L on the lower edge, y = −b, the upper edge being traction free. Find a suitable stress function and the corresponding stress components for this problem, using the weak boundary conditions on x = ±L. 5.3. The beam −b < y < b, 0 < x < L, is built-in at the end x = L and loaded by a linearly-varying compressive normal traction p(x) = Sx/L on the upper edge, y = b,

Problems

81

the remaining edges, x = 0, y = −b being traction free. Find a suitable stress function and the corresponding stress components for this problem, using the weak boundary conditions on x = 0. 5.4. The beam −b < y < b, −L < x < L is simply supported at the ends x = ±L and loaded by a compressive normal traction p(x) = S cos

 πx 2L

on the upper edge, y = b, the lower edge being traction free. Find a suitable stress function and the corresponding stress components for this problem. 5.5. The beam −b < y < b, 0 < x < L, is built-in at the end x = L and loaded by a compressive normal traction p(x) = S sin

 πx 2L

on the upper edge, y = b, the remaining edges, x = 0, y = −b being traction free. Use a combination of the stress function (5.69) and an appropriate polynomial to find the stress components for this problem, using the weak boundary conditions on x = 0. 5.6. The beam −a < x < a, −b < y < b is loaded by a uniform compressive traction p in the central region −a/2 < x < a/2 of both of the edges y = ±b, as shown in Figure 5.5. The remaining edges are traction free. Use a Fourier series with the appropriate symmetries to obtain a solution for the stress field, using the weak condition on σx y on the edges x = ±a and the strong form of all the remaining boundary conditions.

Fig. 5.5

5.7. Use a Fourier series to solve the problem of Figure 5.4(a) in §5.2.3. Choose the terms in the series so as to satisfy the condition σx x (±a, y) = 0 in the strong sense.

82

5 Problems in Rectangular Coördinates

If you are solving this problem in Maple or Mathematica, compare the solution with that of §5.2.3 by making a contour plot of the difference between the truncated Fourier series stress function and the polynomial stress function p  2 3 5 2 2 2 3 2 3 5x . φ= y − y − 15b x y − 5a y + 2b y 40b3 Examine the effect of taking different numbers of terms in the series. 5.8. The large plate y > 0 is loaded at its remote boundaries so as to produce a state of uniform tensile stress σx x = S ; σx y = σ yy = 0 , the boundary y = 0 being traction free. We now wish to determine the perturbation in this simple state of stress that will be produced if the traction-free boundary had a slight waviness, defined by the line y =  cos(λx) , where λ  1. To solve this problem (i) Start with the stress function φ=

Sy 2 + f (y) cos(λx) 2

and determine f (y) if the function φ is to be biharmonic. (ii) The perturbation will be localized near y = 0, so select only those terms in f (y) that decay as y → ∞. (iii) Find the stress components and use the stress-transformation equations to determine the tractions on the wavy boundary. Notice that the inclination of the wavy surface to the plane y = 0 will be everywhere small if λ  1 and hence the trigonometric functions involving this angle can be approximated using sin(z) ≈ z, cos(z) ≈ 1, z  1. (iv) Choose the free constants in f (y) to satisfy the traction-free boundary condition on the wavy surface. (v) Determine the maximum tensile stress and hence the stress-concentration factor as a function of λ. 5.9. A large plate defined by y > 0 is subjected to a sinusoidally varying load σ yy = S sin λx ; σx y = 0 at its plane edge y = 0. Find the complete stress field in the plate and hence estimate the depth y at which the amplitude of the variation in σ yy has fallen to 10% of S. Hint: You might find it easier initially to consider the case of the layer 0 < y < h, with y = h traction free, and then let h → ∞.

Chapter 6

End Effects

The solution of §5.2.2 must be deemed approximate insofar as the boundary conditions on the ends x = ±a of the rectangular beam are satisfied only in the weak sense of force resultants, through equations (5.29–5.31). In general, if a rectangular beam is loaded by tractions of finite polynomial form, a finite polynomial solution can be obtained which satisfies the boundary conditions in the strong (i.e. pointwise) sense on two edges and in the weak sense on the other two edges. The error involved in such an approximation corresponds to the solution of a corrective problem in which the beam is loaded by the difference between the actual tractions applied and those implied by the approximation. These tractions will of course be confined to the edges on which the weak boundary conditions were applied and will be self-equilibrated, since the weak conditions imply that the tractions in the approximate solution have the same force and moment resultants as the actual tractions. For the particular problem of §5.2.2, we note that the stress field of equations (5.54–5.56) satisfies the boundary conditions on the edges y = ±b, but that there is a self-equilibrated normal traction σx x =

p (3b2 y − 5y 3 ) 10b3

(6.1)

on the ends x = ±a, which must be removed by superposing a corrective solution if we wish to satisfy the boundary conditions of Figure 5.3 in the strong sense.

6.1 Decaying Solutions In view of Saint-Venant’s principle, we anticipate that the stresses in the corrective solution will decay as we move away from the edges where the self-equilibrated tractions are applied. The decay rate is likely to be related to the width of the loaded © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 J. R. Barber, Elasticity, Solid Mechanics and Its Applications 172, https://doi.org/10.1007/978-3-031-15214-6_6

83

84

6 End Effects

region and hence we anticipate that the stresses in the corrective solution will be significant only in two regions near the ends, of linear dimensions comparable to the width of the beam. These regions are, shown shaded in Figure 6.1. It follows that the solution of §5.2.2 will be a good approximation in the unshaded region in Figure 6.1.

Fig. 6.1 Regions of the beam influenced by end effects.

It also follows that the corrective solutions for the two ends are uncoupled, since the corrective field for the end x = −a has decayed to negligible proportions before we reach the end x = +a and vice versa. This implies that, so far as the left end is concerned, the corrective solution is essentially identical to that which would be required in the semi-infinite beam, x > −a. We can therefore simplify the statement of the problem by considering the corrective solution for the left end only and shifting the origin to −a, so that the semi-infinite beam under consideration is defined by x > 0, −b < y < b. It is also now clear why we chose to satisfy the strong boundary conditions on the long edges, y = ±b. If instead we had imposed strong conditions on x = ±a and weak conditions on y = ±b, the shaded regions would have overlapped and there would be no region in which the finite polynomial solution was a good approximation to the stresses in the beam. It also follows that the approximation will only be useful when the beam has an aspect ratio significantly different from unity — i.e. a/b  1.

6.2 The Corrective Solution We recall that the stress function φ for the corrective solution must (i) satisfy the biharmonic equation (5.1), (ii) have zero tractions on the boundaries y = ±b — i.e. σ yy = σ yx = 0 ; y = ±b

(6.2)

and (iii) have prescribed non-zero tractions [such as those defined by equation (6.1)] on the end(s), which we can write in the form σx x (0, y) = f 1 (y) ; σx y (0, y) = f 2 (y) ,

(6.3)

6.2 The Corrective Solution

where



b −b

85

 f 1 (y)dy =

b −b

 f 1 (y)ydy =

b

−b

f 2 (y)dy = 0 ,

since the corrective tractions are required to be self-equilibrated. We cannot generally expect to find a solution to satisfy all these conditions in closed form and hence we seek a series or transform (integral) solution as suggested in §5.3. However, the solution will be simpler if we can find a class of solutions, all of which satisfy some of the conditions. We can then write down a more general solution as a superposition of such solutions and choose the coefficients so as to satisfy the remaining condition(s). Since the traction boundary condition on the end will vary from problem to problem, it is convenient to seek solutions which satisfy (i) and (ii) — i.e. biharmonic functions for which ∂2φ ∂2φ = = 0 ; y = ±b . (6.4) ∂x 2 ∂x∂ y

6.2.1 Separated-variable solutions One way to obtain functions satisfying conditions (6.4) is to write them in the separated-variable form φ = f (x)g(y) , (6.5) in which case, (6.4) will be satisfied for all x, provided that g(y) = g  (y) = 0 ; y = ±b .

(6.6)

Notice that the final corrective solution cannot be expected to be of separated-variable form, but we shall see that it can be represented as the sum of a series of such terms. If the functions (6.5) are to be biharmonic, we must have g

d2 f d2g d4g d4 f + 2 + f =0, dx4 d x 2 dy 2 dy 4

(6.7)

and this equation must be satisfied for all values of x, y. Now, if we consider the subset of points (x, c), where c is a constant, it is clear that f (x) must satisfy an equation of the form d4 f d2 f A 4 + B 2 +Cf = 0 , dx dx where A, B, C, are constants, and hence f (x) must consist of exponential terms such as f (x) = exp(λx). Similar considerations apply to the function g(y). Notice

86

6 End Effects

incidentally that λ might be complex or imaginary, giving sinusoidal functions, and there are also degenerate cases where C and/or B = 0 in which case f (x) could also be a polynomial of degree 3 or below. Since we are seeking to represent a field which decays with distance x from the loaded end, we select terms of the form φ = g(y)e−λx , in which case, (6.7) reduces to 2 d4g 2d g + 2λ + λ4 g = 0 , dy 4 dy 2

which is a fourth order ordinary differential equation for g(y) with general solution g(y) = (A1 + A2 y) cos(λy) + (A3 + A4 y) sin(λy) .

(6.8)

6.2.2 The eigenvalue problem The arbitrary constants A1 , A2 , A3 , A4 are determined from the boundary conditions (6.2), which in view of (6.6) lead to the four simultaneous equations (A1 + A2 b) cos(λb) + (A3 + A4 b) sin(λb) = 0

(6.9)

(A1 − A2 b) cos(λb) − (A3 − A4 b) sin(λb) = 0 (A2 + A3 λ + A4 λb) cos(λb) − (A1 λ + A2 λb − A4 ) sin(λb) = 0

(6.10) (6.11)

(A2 + A3 λ − A4 λb) cos(λb) + (A1 λ − A2 λb − A4 ) sin(λb) = 0 . (6.12) This set of equations is homogeneous and will generally have only the trivial solution A1 = A2 = A3 = A4 = 0. However, there are some eigenvalues of the exponential decay rate λ, for which the determinant of coefficients is singular and the solution is non-trivial. A more convenient form of the equations can be obtained by taking sums and differences in pairs — i.e. by constructing the equations (6.9 + 6.10), (6.9 – 6.10), (6.11 + 6.12), (6.11 – 6.12), which after rearrangement and cancellation of non-zero factors yields the set A1 cos(λb) + A4 b sin(λb) = 0

(6.13)

A1 λ sin(λb) − A4 {sin(λb) + λb(cos(λb)} = 0 A2 b cos(λb) + A3 sin(λb) = 0

(6.14) (6.15)

A2 {cos(λb) − λb sin(λb)} + A3 λ cos(λb) = 0 .

(6.16)

What we have done here is to use the symmetry of the system to partition the matrix of coefficients. The terms A1 cos(λy), A4 y sin(λy) are symmetric, whereas

6.2 The Corrective Solution

87

A2 y cos(λy), A3 sin(λy) are antisymmetric. The boundary conditions are also symmetric and hence the symmetric and antisymmetric terms must separately satisfy them. We conclude that the set of equations (6.13–6.16) has two sets of eigenvalues, for one of which the resulting eigenfunctions are symmetric and the other antisymmetric. The symmetric eigenvalues λ(S) are obtained by eliminating A1 , A4 from (6.13, 6.14) with the result (6.17) sin(2λ(S) b) + 2λ(S) b = 0 , whilst the antisymmetric eigenvalues λ(A) are obtained in the same way from (6.15, 6.16) with the result (6.18) sin(2λ(A) b) − 2λ(A) b = 0 . Figure 6.2 demonstrates graphically that the only real solution of equations (6.17, 6.18) is the trivial case λ = 0 (which in fact corresponds to the non-decaying solutions in which an axial force or moment resultant is applied at the end and transmitted along the beam).

Fig. 6.2 Graphical solution of equations (6.17, 6.18).

However, there is a denumerably infinite set of non-trivial complex solutions, corresponding to stress fields which oscillate whilst decaying along the beam. These solutions are fairly easy to find by writing λb = c + ıd, separating real and imaginary parts in the complex equation, and solving the resulting two simultaneous equations for the real numbers, c, d, using a suitable numerical algorithm. Once the eigenvalues have been determined, the corresponding eigenfunctions g (S) , g (A) are readily recovered using (6.8, 6.17, 6.18). We obtain   g (S) = C y sin(λ(S) y) cos(λ(S) b) − b sin(λ(S) b) cos(λ(S) y)   g (A) = D y sin(λ(S) b) cos(λ(S) y) − b sin(λ(S) y) cos(λ(S) b) ,

88

6 End Effects

where C, D are new arbitrary constants related to A1 , A4 and A2 , A3 respectively. We can then establish a more general solution of the form φ=

∞ 

(S)

Ci gi(S) (y)e−λi

x

+

i=1

∞ 

(A)

Di gi(A) (y)e−λi

x

,

(6.19)

i=1

where λi(S) , λi(A) represent the eigenvalues of equations (6.17, 6.18) respectively. The final step is to choose the constants Ci , Di so as to satisfy the prescribed boundary conditions (6.3) on the end x = 0. An improved but still approximate solution of this problem can be obtained by truncating the infinite series at some finite value i = N , defining a scalar measure of the error E in the boundary conditions and then imposing the 2N conditions ∂E =0; ∂Ci

∂E =0 ∂ Di

for

i ∈ [1, N ]

to determine the unknown constants. An appropriate non-negative quadratic error measure is  b  2  2  f 1 (y) − σx x (0, y) + f 2 (y) − σx y (0, y) dy , E= −b

where σx x , σx y are the stress components defined by the truncated series. Since biharmonic boundary-value problems arise in many areas of mechanics, techniques of this kind have received a lot of attention. Convergence of the truncated series is greatly improved if the boundary conditions are continuous in the corners1 . It can be shown that the use of the weak boundary conditions automatically defines continuous values of φ and its first derivatives in the corners (0, −b), (0, b) in the corrective solution. However, convergence problems are still likely to occur if the shear tractions on the two orthogonal edges in the corner are different, for example if (6.20) lim σ yx (x, b) = lim σx y (0, y) . x→0

y→b

Since the tractions on the boundaries are independent, this is a perfectly legitimate physical possibility, but it involves an infinite stress gradient in the corner and leads to a modified Gibbs phenomenon in the series solution, where increase of N leads to greater accuracy over most of the boundary, but to an oscillation of finite amplitude and decreasing wavelength in the immediate vicinity of the corner. The problem can be avoided at the cost of a more complex fundamental problem by extracting a

1

M. I. G. Bloor and M. J. Wilson (2006), An approximate solution method for the biharmonic problem, Proceedings of the Royal Society of London, Vol. A462, pp. 1107–1121.

6.2 The Corrective Solution

89

closed-form solution respresenting the discontinuous tractions. We shall discuss this special solution in §11.1.2 below.

Completeness It is clear that the accuracy of the solution (however defined) can always be improved by taking more terms in the series (6.19) and in particular cases this is easily established numerically. However, it is more challenging to prove that the eigenfunction expansion is complete in the sense that any prescribed self-equilibrated traction on x = 0 can be described to within an arbitrarily small error by taking a sufficient number of terms in the series, though experience with other eigenfunction expansions (e.g. with expansion of elastodynamic states of a structure in terms of normal modes) suggests that this will always be true. In fact, although the analysis described in this section has been known since the investigations by Papkovich and Fadle in the early 1940s, the formal proof of completeness was only completed by Gregory2 in 1980. It is worth noting that, as in many related problems, the eigenfunctions oscillate in y with increasing frequency as λi increases and in fact every time we increase i by 1, an extra zero appears in the function gi (y) in the range 0 < y < b. Thus, there is a certain similarity to the process of approximating functions by Fourier series and in particular, the residual error in case of truncation will always cross zero once more than the last eigenfunction included. This is also helpful in that it enables us to estimate the decay rate of the first excluded term. We see from equation (6.8) that the distance between zeros in y in any of the separate terms would be (π/λ R ), where λ R is the real part of λ. It follows that over a corresponding distance in the x-direction, the field would decay by the factor exp(−π) = 0.0432. This suggests that we might estimate the decay rate of the end field by noting the distance between zeros in the corresponding tractions3 . For example, the traction of equation (6.1) has zeros at y = 0, 3b/5, corresponding to λ R = (5π/3b) = 5.23/b. An alternative way of estimating the decay rate is to note that the decay rate for the various terms in (6.19) increases with i and hence as x increases, the leading term will tend to predominate. The tractions of (6.1) are antisymmetric and hence

2

R. D. Gregory (1979), Green’s functions, bi-linear forms and completeness of the eigenfunctions for the elastostatic strip and wedge, J. Elasticity, Vol. 9, pp. 283–309; R. D. Gregory (1980), The semi-infinite strip x ≥ 0, −1 ≤ y ≤ 1; completeness of the Papkovich-Fadle eigenfunctions when φx x (0, y), φ yy (0, y) are prescribed, J. Elasticity, Vol. 10, pp. 57–80; R. D. Gregory (1980), The traction boundary-value problem for the elastostatic semi-infinite strip; existence of solution and completeness of the Papkovich-Fadle eigenfunctions, J. Elasticity, Vol. 10, pp. 295–327. These papers also include extensive references to earlier investigations of the problem. 3 This assumes that the wavelength of the tractions is the same as that of φ, which of course is an approximation, since neither function is purely sinusoidal.

90

6 End Effects

the leading term corresponds to the real part of the first eigenvalue of equation (6.18), which is found numerically to be λ R b = 3.7. Either way, we can conclude that the error associated with the end tractions in the approximate solution of §5.2.2 has decayed to around e−4 , i.e. to about 2% of the values at the end, within a distance b of the end. Thus the region affected by the end condition — the shaded region in Figure 6.1 — is quite small. For problems which are symmetric in y, the leading self-equilibrated term is likely to have the form of Figure 6.3(a), which has a longer wavelength than the corresponding antisymmetric form, 6.3(b). The end effects in symmetric problems therefore decay more slowly and this is confirmed by the fact that the real part of the first eigenvalue of the symmetric equation (6.17) is only λ R b = 2.1.

Fig. 6.3 Leading term in self-equilibrated tractions for (a) symmetric loading and (b) antisymmetric loading.

6.3 Other Saint-Venant Problems The general strategy used in §6.2 can be applied to other curvilinear coördinate systems to correct the errors incurred by imposing the weak boundary conditions on appropriate edges. The essential steps are:(i) Define a coördinate system (ξ, η) such that the boundaries on which the strong conditions are applied are of the form, η = constant. (ii) Find a class of separated-variable biharmonic functions containing a parameter (λ in the above case). (iii) Set up a system of four homogeneous equations for the coefficients of each function, based on the four traction-free boundary conditions for the corrective solution on the edges η = constant. (iv) Find the eigenvalues of the parameter for which the system has a non-trivial solution and the corresponding eigenfunctions, which are then used as the terms in an eigenfunction expansion to define a general form for the corrective field.

6.4 Mathieu’s Solution

91

(v) Determine the coefficients in the eigenfunction expansion from the prescribed inhomogeneous boundary conditions on the end ξ = constant.

6.4 Mathieu’s Solution The method described in §6.2 is particularly suitable for rectangular bodies of relatively large aspect ratio, since the weak solution is then quite accurate over a substantial part of the domain and the corrections at the two ends are essentially independent of each other. In other cases, and particularly for the square b = a, we lose these advantages and the inconvenience of solving the eigenvalue equations (6.17, 6.18) tilts the balance in favour of an alternative method due originally to Mathieu, who represented the solution of the entire problem as the sum of two orthogonal Fourier series of the form (5.84). The general problem can be decomposed into four sub-problems which are respectively either symmetric or antisymmetric with respect to the x- and y-axes. Following Meleshko and Gomilko4 , we restrict attention to the symmetric/symmetric case, defined by the boundary conditions σx x (a, y) = σx x (−a, y) = f (y) ; σx y (a, y) = −σx y (−a, y) = g(y) (6.21) σ yx (x, b) = −σ yx (x, −b) = h(x) ; σ yy (x, b) = σ yy (x, −b) = (x) , (6.22) where f,  are even and g, h odd functions of their respective variables. We shall also assume that there is no discontinuity in shear traction of the form (6.20) at the corners and hence that g(b) = h(a). We define a biharmonic stress function with the appropriate symmetries as φ = A0 y 2 +

∞ 

(Am y sinh(αm y) + Cm cosh(αm y)) cos(αm x)

m=1 ∞ 

+B0 x 2 +

(Bn x sinh(βn x) + Dn cosh(βn x)) cos(βn y) .

(6.23)

n=1

As in §5.3.1, the αm , βn can be chosen so as to ensure that each term gives either zero shear tractions or zero normal tractions on two opposite edges, but not both. For example, with mπ nπ ; βn = , (6.24) αm = a b 4

V. V. Meleshko and A. M. Gomilko (1997), Infinite systems for a biharmonic problem in a rectangle, Proceedings of the Royal Society of London, Vol. A453, pp. 2139–2160. The reader should be warned that there are several typographical errors in this paper, some of which are corrected in V. V. Meleshko and A. M. Gomilko (2004), Infinite systems for a biharmonic problem in a rectangle: further discussion, Proceeedings of the Royal Society of London, Vol. A460, pp. 807–819.

92

6 End Effects

the second series makes no contribution to the shear traction on y = b, whilst the first series (that involving the coefficients Am , Cm ) makes no contribution to the shear traction on x = a. The shear traction on x = a is then given by σx y (a, y) =



 1 + Dn sinh(βn a) sin(βn y) , βn2 Bn a coth(βn a) + βn n=1

∞ 

using (4.6), and substituting into (6.21)2 and applying the Fourier inversion theorem5 , we have

 b 1 1 + Dn = 2 g(y) sin(βn y)dy , Bn a coth(βn a) + βn βn b sinh(βn a) −b which defines a linear relation between each individual pair of constants Bn , Dn . A similar relation between Am , Cm can be obtained from (6.22)1 and these conditions can then be used to eliminate Cm , Dn in (6.23), giving φ = A0 y 2 −

∞ 

Am B(y, αm , b) sinh(αm b) cos(αm x)

m=1 ∞ 

+B0 x 2 −

Bn B(x, βn , a) sinh(βn a) cos(βn y)

n=1

+

∞ ∞  gn cosh(βn x) cos(βn y)  h m cosh(αm y) cos(αm x) + , 2 a sinh(α b) βn2 b sinh(βn a) αm m n=1 m=1



where gn = and

b

−b

 g(y) sin(βn y)dy ;

hm =

a

−a

h(x) sin(αm x)d x ,

sinh(λz) 1 cosh(λz) −z . B(z, λ, h) = h coth(λh) + λ sinh(λh) sinh(λh)

The normal stress components are then obtained as σx x = 2 A0 −

∞ 

2 Am αm A(y, αm , b) sinh(αm b) cos(αm x)

m=1

+

∞ 

Bn βn2 B(x, βn , a) sinh(βn a) cos(βn y)

n=1 ∞ ∞  gn cosh(βn x) cos(βn y)  h m cosh(αm y) cos(αm x) + , − b sinh(βn a) a sinh(αm b) n=1 m=1

5

See for example equation (5.89).

6.4 Mathieu’s Solution

σ yy = 2B0 +

93

∞ 

2 Am αm B(y, αm , b) sinh(αm b) cos(αm x)

m=1

− +

∞  n=1 ∞  n=1

where

Bn βn2 A(x, βn , a) sinh(βn a) cos(βn y) ∞

gn cosh(βn x) cos(βn y)  h m cosh(αm y) cos(αm x) − , b sinh(βn a) a sinh(αm b) m=1



1 cosh(λz) sinh(λz) A(z, λ, h) = h coth(λh) − −z . λ sinh(λh) sinh(λh)

Imposing the remaining two boundary conditions (6.21)1 , (6.22)2 and applying the Fourier inversion theorem to the resulting equations, we then obtain the infinite set of simultaneous equations X m bB(b, αm , b) =

∞  n=1

Yn a B(a, βn , a) =

2 4Yn αm + Hm , 2 + β 2 )2 (αm n

∞ 

4X m βn2 + Kn , 2 + β 2 )2 (αm n m=1

for m, n ≥ 1, where Xm =

2 sinh(αm b) (−1)m+1 Am αm ; b

Yn =

(−1)n Bn βn2 sinh(βn a) a ∞

Hm =

(−1)m+1 [m + coth(αm b)h m ]  2(−1)n βn gn + 2 + β 2 )ab a (αm n n=1

Kn =

(−1)n [ f n + coth(βn a)gn ]  2(−1)m αm h m − 2 + β 2 )ba b (αm n m=1

∞



and fn =

b −b

 f (y) cos(βn y)dy ; m =

a −a

(x) cos(αm x)d x .

Also, the zeroth-order Fourier inversion yields the constants A0 , B0 as ∞

A0 =

f 0  (−1)m h m − ; 4b m=1 2αm ba

∞

B0 =

0  (−1)n gn − . 4a n=1 2βn ab

With this solution, the coefficients X m , Yn generally tend to a common constant value G at large m, n and the series solution exhibits a non-vanishing error near the

94

6 End Effects

corners of the rectangle6 . A more convergent solution can be obtained by defining new constants through the relations X m = X˜ m + G ;

Yn = Y˜n + G .

so that X˜ m , Y˜n decay with increasing m, n. The terms involving G can then be summed explicitly yielding the additional polynomial term φ0 =

 G  2 (y − b2 )2 − (x 2 − a 2 )2 24

(6.25)

in the stress function φ. Unfortunately, the required value of G cannot be obtained in closed form except in certain special cases. However, an approximation can be obtained using the Rayleigh-Ritz method7 , leading to the result  b  

   1 a 1 45 b2 a2 2 2 dy − dx . G= f (y) y − (x) x − 4(a 4 + b4 ) b −b 3 a −a 3 (6.26) This agrees with the exact value cited by Meleshko and Gomilko for the special case of the square b = a with (x) a quadratic function of x.

Alternative series solutions In the technique described above, the parameters αm , βn were chosen so as to simplify the satisfaction of the shear traction boundary conditions, after which the normal traction conditions led to an infinite set of algebraic equations. An alternative approach is to reverse this procedure by choosing αm =

(2m − 1)π (2n − 1)π ; βn = 2a 2b

(6.27)

and omitting the terms with A0 , B0 . Equation (6.23) then defines a stress field in which the first series makes no contribution to the normal tractions on x = a, and the second series makes no contribution to the normal tractions on y = b. These boundary conditions imply a one-to-one relation between Am , Cm , after which the shear traction boundary conditions lead to an infinite set of algebraic equations. One advantage of this version is that it leads to a convergent solution even in the case where g(b) = h(a) and hence the shear tractions are discontinuous in the corners8 , though convergence is slow in such cases. An alternative method of treating this

6

Meleshko and Gomilko loc. cit. See §37.5 and Problem 37.5 below. 8 Meleshko and Gomilko loc. cit. 7

Problems

95

discontinuity is to extract it explicitly, as discussed in §11.1.2 below, and then use the ‘even’ series defined by equations (6.24) for the corrective problem.

Problems 6.1. Show that if ζ = x +ı y and sin(ζ)−ζ = 0, where x, y are real variables, then

 f (x) ≡ cos x x 2 − sin2 x + sin x ln(sin x) − sin x ln x + x 2 − sin2 x = 0 . Using Maple or Mathematica to plot the function f (x), find the first six roots of this equation and hence determine the first six values of λ R b for the antisymmetric mode. 6.2. Devise a method similar to that outlined in Problem 6.1 to determine the first six values of λ R b for the symmetric mode. 6.3. A displacement function representation for plane strain problems can be developed9 in terms of two harmonic functions φ, ω in the form 2μu x = σx x =

∂ω ∂ω ∂φ ∂φ +y ; 2μu y = +y − (3 − 4ν)ω ∂x ∂x ∂y ∂y

∂2ω ∂ω ∂2φ ∂2ω ∂ω ∂2φ ; σ + y − (1 − 2ν) + y − 2ν = x y 2 2 ∂x ∂x ∂y ∂x∂ y ∂x∂ y ∂x σ yy =

∂2φ ∂2ω ∂ω + y 2 − 2(1 − ν) . 2 ∂y ∂y ∂y

Use this representation to formulate the eigenvalue problem of the long strip x > 0, −b < y < b whose edges y = ±b are both bonded to a rigid body. Find the eigenvalue equation for symmetric and antisymmetric modes and comment on the expected decay rates for loading of the strip on the end x = 0. 6.4. Use the displacement function representation of Problem 6.3 to formulate the eigenvalue problem for the long strip x > 0, −b < y < b whose edges y = ±b are in frictionless contact with a rigid body (so that the normal displacement is zero, but the frictional (tangential) traction is zero). Find the eigenvalue equation for symmetric and antisymmetric modes and comment on the expected decay rates for loading of the strip on the end x = 0. 6.5. Use the displacement function representation of Problem 6.3 to formulate the eigenvalue problem for the long strip x > 0, −b < y < b which is bonded to a rigid surface at y = −b, the other long edge y = b being traction free. Notice that this 9

See §22.5.3.

96

6 End Effects

problem is not symmetrical, so the problem will not partition into symmetric and antisymmetric modes. 6.6. Use Mathieu’s method to approximate the stresses in the square −a < x < a, −a < y < a subjected to the tractions σx x (a, y) = σx x (−a, y) =

Sy 2 , a2

all the remaining tractions being zero. Start by using (6.25, 6.26) to define a polynomial first approximation to the solution. Then use the series (6.23) to define a corrective solution — i.e. the stress function which when added to φ0 defines the complete solution. The constants X m , Yn in the corrective solution will then decay with increasing m, n and a good approximation can be found by truncating the series at m = n = 2.

Chapter 7

Body Forces

A body force is defined as one which acts directly on the interior particles of the body, rather than on the boundary. Since the interior of the body is not accessible, it follows necessarily that body forces can only be produced by some kind of physical process which acts ‘at a distance’. The commonest examples are forces due to gravity and magnetic or electrostatic attraction. In addition, we can formulate quasi-static elasticity problems for accelerating bodies in terms of body forces, using d’Alembert’s principle (see §7.2.2 below).

7.1 Stress Function Formulation We noted in §4.2.2 that the Airy stress function formulation satisfies the equilibrium equations if and only if the body forces are identically zero, but the method can be extended to the case of non-zero body forces provided the latter can be expressed as the gradient of a scalar potential, V . We adopt the new definitions ∂2φ +V ∂ y2 ∂2φ σ yy = +V ∂x 2 ∂2φ , σx y = = − ∂x∂ y σx x =

(7.1) (7.2) (7.3)

Supplementary Information The online version contains supplementary material available at https://doi.org/10.1007/978-3-031-15214-6_7.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 J. R. Barber, Elasticity, Solid Mechanics and Its Applications 172, https://doi.org/10.1007/978-3-031-15214-6_7

97

98

7 Body Forces

in which case the two-dimensional equilibrium equations will be satisfied if and only if ∂V ∂V ; py = − , (7.4) px = − ∂x ∂y i.e.

p = −∇V .

(7.5)

Notice that the body force potential V appears in the definitions of the two normal stress components σx x , σ yy (7.1, 7.2), but not in the shear stress σx y (7.3). Thus, the modification in these equations is equivalent to the addition of a biaxial hydrostatic tension of magnitude V , which is of course invariant under coördinate transformation (see §1.1.4).

7.1.1 Conservative vector fields If a force field is capable of being represented as the gradient of a scalar potential, as in equation (7.5), it is referred to as conservative. This terminology arises from gravitational theory, since if we move a particle around in a gravitational field and if the force on the particle varies with position, the principle of conservation of energy demands that the work done in moving the particle from point A to point B should be path-independent — or equivalently, the work done in moving it around a closed path should be zero. If this were not the case, we could choose a direction for the particle to move around the path which would release energy and hence have an inexhaustible source of energy. If all such integrals are path-independent, we can use the work done in bringing a particle from infinity to a given point as the definition of a unique local potential. Then, by equating the work done in an infinitesimal motion to the corresponding change in potential energy, we can show that the local force is proportional to the gradient of the potential, thus demonstrating that a conservative force field must be capable of a representation like (7.5). Conversely, if a given force field can be represented in this form, we can show by integration that the work done in moving a particle from A to B is proportional to V (A)−V (B) and is therefore path-independent. Not all body force fields are conservative and hence the formulation of §7.1 is not sufficiently general for all problems. However, we shall show below that most of the important problems involving body forces can be so treated. We can develop a condition for a vector field to be conservative in the same way as we developed the compatibility conditions for strains. We argue that the two independent body force components px , p y are defined in terms of a single scalar potential V and hence we can obtain a constraint equation on px , p y by eliminating V between equations (7.4) with the result ∂ py ∂ px − =0. ∂x ∂y

(7.6)

7.2 Particular Cases

99

In three dimensions there are three equations like (7.6), from which we conclude that a vector field p is conservative if and only if curl p = 0. Another name for such fields is irrotational, since we note that if we replace p by the displacement vector u, the conditions like (7.6) are equivalent to the statement that the rotation ω is identically zero [cf equation (1.33)]. If the body force field satisfies equation (7.6), the corresponding potential can be recovered by partial integration. We shall illustrate this procedure in §7.2 below.

7.1.2 The compatibility condition We demonstrated in §4.3.1 that, in the absence of body forces, the compatibility condition reduces to the requirement that the Airy stress function φ be biharmonic. This condition is modified when body forces are present. We follow the same procedure as in §4.3.1, but use equations (7.1–7.3) in place of (4.6). Substituting into the compatibility equation in terms of stresses (4.7), we obtain ∂4φ ∂2 V ∂4φ ∂2 V ∂4φ + − ν − ν + 2(1 + ν) ∂ y4 ∂ y2 ∂x 2 ∂ y 2 ∂ y2 ∂x 2 ∂ y 2 4 2 4 ∂ φ ∂ V ∂ φ ∂2 V + 4 + − ν − ν =0, ∂x ∂x 2 ∂x 2 ∂ y 2 ∂x 2 i.e. ∇ 4 φ = −(1 − ν)∇ 2 V .

(7.7)

Methods of obtaining suitable functions which satisfy this equation will be discussed in §7.3 below.

7.2 Particular Cases It is worth noting that the vast majority of mechanical engineering components are loaded principally by boundary tractions rather than body forces. Of course, most components are subject to gravity loading, but the boundary loads are generally so much larger than sefl-weight that gravity can be neglected. This is less true for civil engineering structures such as buildings, where the self-weight of the structure may be much larger than the weight of the contents or wind loads, but even in this case it is important to distinguish between the gravity loading on the individual component and that transmitted to the component from other parts of the structure by way of boundary tractions. It might be instructive at this point for the reader to draw free-body diagrams for a few common engineering components and identify the sources and relative

100

7 Body Forces

magnitudes of the forces acting upon them. There are really comparatively few ways of applying a load to a body. By far the commonest is to push against it with another body — in other words to apply the load by contact. This is why contact problems occupy a central place in elasticity theory1 . Significant loads may also be applied by fluid pressure as in the case of turbine blades or aircraft wings. Notice that it is fairly easy to apply a compressive normal traction to a boundary, but much harder to apply tension or shear. This preamble might be taken as a justification for not studying the subject of body forces at all, but there are a few applications in which they are of critical importance, most notably those dynamic problems in which large accelerations occur. We shall develop expressions for the body force potential for some important cases in the following sections.

7.2.1 Gravitational loading The simplest type of body force loading is that due to a gravitational field. If the problem is to remain two-dimensional, the direction of the gravitational force must lie in the x y-plane and we can choose it to be in the negative y-direction without loss of generality. The magnitude of the force will be ρg per unit volume, where ρ is the density of the material, so in the notation of §2.1 we have px = 0 ; p y = −ρg . This force field clearly satisfies condition (7.6) and is therefore conservative, and by inspection we note that it can be derived from the body force potential V = ρgy . It also follows that ∇ 2 V = 0 and hence the stress function, φ for problems involving gravitational loading is biharmonic, from (7.6).

7.2.2 Inertia forces It might be argued that Jean d’Alembert achieved immortality simply by moving a term from one side of an equation to the other, since d’Alembert’s principle consists merely of writing Newton’s second law of motion in the form F − ma = 0 and treating the term −ma as a fictitious force in order to reduce the dynamic problem to one in statics. This simple process enables us to formulate elasticity problems for accelerating bodies as elastostatic body force problems, the corresponding body forces being 1

See Chapters 12, 31, 34.

7.2 Particular Cases

101

px = −ρax ; p y = −ρa y ,

(7.8)

where ax , a y are the local components of acceleration, which may of course vary with position through the body.

7.2.3 Quasi-static problems If the body were rigid, the accelerations of equation (7.8) would be restricted to those associated with rigid-body translation and rotation, but in a deformable body, the distance between two points can change, giving rise to additional, stress-dependent terms in the accelerations. These two effects give qualitatively distinct behaviour, both mathematically and physically. In the former case, the accelerations will generally be defined a priori from the kinematics of the problem, which therefore reduces to an elasticity problem with body forces. We shall refer to such problems as quasi-static. By contrast, when the accelerations associated with deformations are important, they are not known a priori, since the stresses producing the deformations are themselves part of the solution. In this case the kinematic and elastic problems are coupled and must be solved together. The resulting equations are those governing the propagation of elastic waves through a solid body and their study is known as Elastodynamics. In this chapter, we shall restrict attention to quasi-static problems. As a practical point, we note that the characteristic time scale of elastodynamic problems is very short. For example it generally takes only a very short time for an elastic wave to traverse a solid. If the applied loads are applied gradually in comparison with this time scale, the quasi-static assumption generally gives good results. A measure of the success of this approximation is that it works quite well even for the case of elastic impact between bodies, which may have a duration of the order of a few milliseconds2 .

7.2.4 Rigid-body kinematics The most general acceleration for a rigid body in the plane involves arbitrary translation and rotation. We choose a coördinate system fixed in the body and suppose that,

2

For more information about elastodynamic problems, the reader is referred to the classical texts of J. D. Achenbach, Wave Propagation in Elastic Solids, North Holland, Amsterdam, 1973 and A. C. Eringen and E. S. Suhubi, ¸ Elastodynamics, Academic Press, New York, 1975. For a more detailed discussion of the impact of elastic bodies, see J. R. Barber, Contact Mechanics, Springer, Cham, 2018, Chapter 20, K. L. Johnson, Contact Mechanics, Cambridge University Press, Cambridge, 1985, §11.4, W. Goldsmith, Impact, Arnold, London, 1960.

102

7 Body Forces

at some instant, the origin has velocity v0 and acceleration a0 and the body is rotating ˙ in the clockwise sense with absolute angular velocity Ω and angular acceleration Ω. The instantaneous acceleration of the point (r, θ) relative to the origin can then be written ˙ . ar = −Ω 2 r ; aθ = −Ωr Transforming these results into the x, y-coördinate system and adding the acceleration of the origin, we obtain the components of acceleration of the point (x, y) as ax = a0x − Ω 2 x + Ω˙ y a y = a0y − Ω 2 y − Ω˙ x

(7.9) (7.10)

and hence the corresponding body force field is px = −ρ(a0x − Ω 2 x + Ω˙ y) p y = −ρ(a0y − Ω 2 y − Ω˙ x) .

(7.11) (7.12)

The astute reader will notice that the case of gravitational loading can be recovered as a special case of these results by writing a0y = g and setting all the other terms to zero. In fact, a reasonable interpretation of the gravitational force is as a d’Alembert force consequent on resisting the gravitational acceleration. Notice that if a body is in free fall — i.e. if it is accelerating freely in a gravitational field and not rotating — there is no body force and hence no internal stress unless the boundaries are loaded. Substitution of (7.11, 7.12) into (7.6) shows that the inertia forces due to rigid-body accelerations are conservative if and only if Ω˙ = 0 — i.e. if the angular velocity is constant. We shall determine the body force potential for this special case. Methods of treating the problem with non-zero angular acceleration are discussed in §7.4 below. From equations (7.4, 7.11, 7.12) with Ω˙ = 0 we have ∂V = ρ(a0x − Ω 2 x) ∂x ∂V = ρ(a0y − Ω 2 y) , ∂y

(7.13) (7.14)

and hence, on partial integration of (7.13)   1 V = ρ a0x x − Ω 2 x 2 + h(y) , 2 where h(y) is an arbitrary function of y only. Substituting this result into (7.14) we obtain the ordinary differential equation

7.3 Solution for the Stress Function

103

dh = ρ(a0y − Ω 2 y) dy

(7.15)

for h(y), which has the general solution   Ω 2 y2 +C , h(y) = ρ a0y y − 2 where C is an arbitrary constant which can be taken to be zero without loss of generality, since we are only seeking a particular potential function V . The final expression for V is therefore   1 2 2 2 V = ρ a0x x + a0y y − Ω (x + y ) . 2

(7.16)

The reader might like to try this procedure on a set of body forces which do not satisfy the condition (7.6). It will be found that the right-hand side of the ordinary differential equation like (7.15) then contains terms which depend on x and hence this equation cannot be solved for h(y).

7.3 Solution for the Stress Function Once the body force potential V has been determined, the next step is to find a suitable function φ, which satisfies the compatibility condition (7.7) and which defines stresses through equations (7.1–7.3) satisfying the boundary conditions of the problem. There are broadly speaking two ways of doing this. One is to choose some suitable form (such as a polynomial) without regard to equation (7.7) and then satisfy the constraint conditions resulting from (7.7) in the same step as those arising from the boundary conditions. The other is to seek a general solution of the inhomogeneous equation (7.7) and then determine the resulting arbitrary constants from the boundary conditions.

7.3.1 The rotating rectangular bar As an illustration of the first method, we consider the problem of the rectangular bar −a < x < a, −b < y < b, rotating about the origin at constant angular velocity Ω, all the boundaries being traction free (see Figure 7.1). The body force potential for this problem is obtained from equation (7.16) as 1 V = − ρΩ 2 (x 2 + y 2 ) , 2 and hence the stress function must satisfy the equation

(7.17)

104

7 Body Forces

Fig. 7.1 The rotating rectangular bar.

∇ 4 φ = 2ρ(1 − ν)Ω 2 ,

(7.18)

from (7.7, 7.17). The geometry suggests a formulation in Cartesian coördinates and equation (7.18) leads us to expect a polynomial of degree 4 in x, y. We also note that V is even in both x and y and that the boundary conditions are homogeneous, so we propose the candidate stress function φ = A1 x 4 + A2 x 2 y 2 + A3 y 4 + A4 x 2 + A5 y 2 ,

(7.19)

which contains all the terms of degree 4 and below with the required symmetry. The constants A1 , . . . , A5 will be determined from equation (7.18) and from the boundary conditions



b −b

σ yx = 0 ; y = ±b σ yy = 0 ; y = ±b

(7.20) (7.21)

σx x dy = 0 ; x = ±a ,

(7.22)

where we have applied the weak boundary conditions only on x = ±a. Note also that the other two weak boundary conditions — that there should be no moment and no shear force on the ends — are satisfied identically in view of the symmetry of the problem about y = 0. As in the problem of §5.2.3, it is algebraically simpler to start by satisfying the strong boundary conditions (7.20, 7.21). Substituting (7.19) into (7.1–7.3), we obtain 1 σx x = 2 A2 x 2 + 12 A3 y 2 + 2 A5 − ρΩ 2 (x 2 + y 2 ) 2 1 σ yy = 12 A1 x 2 + 2 A2 y 2 + 2 A4 − ρΩ 2 (x 2 + y 2 ) 2 σ yx = −4 A2 x y .

(7.23) (7.24) (7.25)

7.3 Solution for the Stress Function

105

It follows that conditions (7.20, 7.21) will be satisfied for all x if and only if A1 = ρΩ 2 /24 ; A2 = 0 ; A4 = ρΩ 2 b2 /4 .

(7.26)

The constant A3 can now be determined by substituting (7.19) into (7.18), with the result 24(A1 + A3 ) = 2ρΩ 2 (1 − ν) , which is the inhomogeneous equivalent of the constraint equations (see §5.1), and hence A3 = ρΩ 2 (1 − 2ν)/24 , using (7.26). Finally, we determine the remaining constant A5 by substituting (7.23) into the weak boundary condition (7.22) and evaluating the integral, with the result  2  νb a2 + . A5 = ρΩ 2 6 4 The final stress field is therefore  2  2 ν(b2 − 3y 2 ) 2 (a − x ) + ; σx x = ρΩ 2 3

σ yy =

ρΩ 2 2 (b − y 2 ) 2

σ yx = 0 , from equations (7.23–7.25). Notice that the boundary conditions on the ends x = ±a agree with those of the physical problem except for the second term in σx x , which represents a symmetric self-equilibrated traction. From §6.2.2, we anticipate that the error due to this disagreement will be confined to regions near the ends of length comparable with the half-length, b.

7.3.2 Solution of the governing equation Equation (7.18) is an inhomogeneous partial differential equation — i.e. it has a known function of x, y on the right-hand side — and it can be solved in the same way as an inhomogeneous ordinary differential equation, by first finding a particular solution of the equation and then superposing the general solution of the corresponding homogeneous equation. In this context, a particular solution is any function φ that satisfies (7.18). It contains no arbitrary constants. The generality in the general solution comes from arbitrary constants in the homogeneous solution. Furthermore, the homogeneous solution is the solution of equation (7.18) modified to make the right-hand side zero — i.e.

106

7 Body Forces

∇ 4φ = 0 Thus, the homogeneous solution is a general biharmonic function and we can summarize the solution method as containing the three steps:(i) finding any function φ which satisfies (7.18); (ii) superposing a sufficiently general biharmonic function (which therefore contains several arbitrary constants); (iii) choosing the arbitrary constants to satisfy the boundary conditions. In the problem of the preceding section, the particular solution would be any fourth degree polynomial satisfying (7.18) and the homogeneous solution, the most general fourth degree biharmonic function with the appropriate symmetry. Notice that the particular solution (i) is itself a solution of a different physical problem — one in which the body forces are correctly represented, but the correct boundary conditions are not (usually) satisfied. Thus, the function of the homogeneous solution is to introduce additional degrees of freedom to enable us to satisfy the boundary conditions.

Physical superposition It is often helpful to think of this superposition process in a physical rather than a mathematical sense. In other words, we devise a related problem in which the body forces are the same as in the real problem, but where the boundary conditions are simpler. For example, if a beam is subjected to gravitational loading, a simple physical ‘particular’ solution would correspond to the problem in which the beam is resting on a rigid foundation and hence the stress field is one-dimensional. To complete the real problem, we would then have to superpose the solution of a corrective problem with tractions equal to the difference between the required tractions and those implied by the particular solution, but with no body force, since this has already been taken into account in the particular solution. One advantage of this way of thinking is that it is not restricted to problems in which the body force can be represented by a potential. We can therefore use it to solve the problem of the rotationally accelerating beam in the next section.

7.4 Rotational Acceleration We saw in §7.2.4 that the body force potential cannot be used in problems where the angular acceleration is non-zero. In this section, we shall generate a particular

7.4 Rotational Acceleration

107

solution for this problem and then generalize it, using the results without body force from Chapter 5.

7.4.1 The circular disk Consider the rotationally symmetric problem of Figure 7.2. A solid circular disk, radius a, is initially at rest (Ω = 0) and at time t = 0 it is caused to accelerate in a clockwise direction with angular acceleration Ω˙ by shear tractions S uniformly distributed around the edge r = a. Note that we could determine the magnitude of these tractions by writing the equation of motion for the disk, but it will not be necessary to do this — the result will emerge from the analysis of the stress field.

Fig. 7.2 The disk with rotational acceleration.

At any given time, the body forces and hence the resulting stress field will be the sum of two parts, one due to the instantaneous angular velocity and the other to the angular acceleration. The former can be obtained using the body force potential and we therefore concentrate here on the contribution due to the angular acceleration. This is the only body force at the beginning of the process, so the following solution can also be regarded as the solution for the instant, t = 0. The problem is clearly axisymmetric, so the stresses and displacements must depend on the radius r only, but it is also antisymmetric, since if the sign of the tractions were changed, the problem would become a mirror image of that illustrated. Now suppose some point in the disk had a non-zero outward radial displacement, u r . Changing the sign of the tractions would change the sign of this displacement — i.e. make it be directed inwards — but this is impossible if the new problem is to be a mirror image of the old. We can therefore conclude from symmetry that u r = 0 throughout the disk. In the same way, we can conclude that the stress components

108

7 Body Forces

σrr , σθθ are zero everywhere. Incidentally, we also note that if the problem is conceived as being one of plane strain, symmetry demands that σzz be everywhere zero. It follows that this is one of those lucky problems in which plane stress and plane strain are the same and hence the plane stress assumption involves no approximation. We conclude that there is only one non-zero displacement component, u θ , and one non-zero strain component, er θ . Thus, the number of strain and displacement components is equal and no non-trivial compatibility conditions can be obtained by eliminating the displacement components. (An alternative statement is that all the compatibility equations are satisfied identically, as can be verified by substitution — the compatibility equations in cylindrical polar coördinates are given by Saada3 ). It follows that the only non-zero stress component, σr θ can be determined from equilibrium considerations alone. Considering the equilibrium of a small element due to forces in the θ-direction and dropping terms which are zero due to the symmetry of the system, we obtain dσr θ 2σr θ + = −ρr Ω˙ , (7.27) dr r which has the general solution σr θ = −

˙ 2 A ρΩr + 2 , 4 r

where A is an arbitrary constant which must be set to zero to retain continuity at the origin. Thus, the stress field in the disk of Figure 7.2 is σr θ = −

˙ 2 ρΩr . 4

(7.28)

In particular, the traction at the surface r = a is S = −σr θ (a) =

˙ 2 ρΩa 4

and the applied moment about the axis of rotation is therefore M = 2πa 2 S =

˙ 4 πρΩa , 2

(7.29)

since the traction acts over a length 2πa and the moment arm is a. We know from ˙ where I is the moment of inertia and we can elementary dynamics that M = I Ω, therefore deduce from (7.29) that I = 3

πρa 4 , 2

A. S. Saada, Elasticity, Pergamon Press, New York, 1973, §6.9.

7.4 Rotational Acceleration

109

which is of course the correct expression for the moment of inertia of a solid disk of uniform density ρ, radius a and unit thickness. Notice however that we were able to deduce the relation between M and Ω˙ without using the equations of rigid-body dynamics. Equation (7.27) ensures that every particle of the body obeys Newton’s second law, and this is sufficient to ensure that the complete body satisfies the equations of rigid-body dynamics.

7.4.2 The rectangular bar We can use the stress field (7.28) as a particular solution for determining the stresses in a body of any shape due to angular acceleration. One way to think of this is to imagine cutting out the real body from an imaginary disk of sufficiently large radius. The stresses in the cut-out body will be the same as those in the disk, provided we arrange to apply tractions to the boundaries of the body that are equal to the stress components on those surfaces before the cut was made. These will not generally be the correct boundary tractions for the real problem, but we can correct the boundary tractions by superposing a homogeneous solution — i.e. a corrective solution for the actual body with prescribed boundary tractions (equal to the difference between those actually applied and those implied by the disk solution) but no body forces (since these have already been taken care of in the disk solution). As an illustration, we consider the rectangular bar, −a < x < a, −b < y < b, accelerated by two equal shear forces, F, applied at the ends x = ±a as shown in Figure 7.3, the other boundaries, y = ±b being traction free.

Fig. 7.3 The rectangular bar with rotational acceleration.

We first transform the disk solution into rectangular coördinates, using the transformation relations (1.9–1.11), obtaining

110

7 Body Forces

˙ 2 sin θ cos θ ρΩr ρΩ˙ x y = (7.30) 2 2 ˙ 2 sin θ cos θ ρΩ˙ x y ρΩr =− (7.31) = 2σr θ sin θ cos θ = − 2 2 ˙ 2 − y2) ˙ 2 (cos2 θ − sin2 θ) ρΩ(x ρΩr =− . = σr θ (cos2 θ − sin2 θ) = − 2 4 (7.32)

σx x = −2σr θ sin θ cos θ = σ yy σx y

This stress field is clearly odd in both x and y and involves normal tractions on y = ±b which vary linearly with x. The bending moment will therefore vary with x 3 suggesting a stress function φ with a leading term x 5 y — i.e. a sixth degree polynomial. The most general polynomial with the appropriate symmetry is φ = A1 x 5 y + A2 x 3 y 3 + A3 x y 5 + A4 x 3 y + A5 x y 3 + A6 x y .

(7.33)

The stress components are the sum of those obtained from the homogeneous stress function (7.33) using the definitions (4.6) — remember the homogeneous solution here is one without body force — and those from the disk problem given by equations (7.30–7.32). We find 1 ρΩ˙ x y + 6A2 x 3 y + 20 A3 x y 3 + 6A5 x y 2 1 = − ρΩ˙ x y + 20 A1 x 3 y + 6A2 x y 3 + 6A4 x y 2 1 ˙ 2 − y 2 ) − 5A1 x 4 − 9A2 x 2 y 2 = − ρΩ(x 4 −5A3 y 4 − 3A4 x 2 − 3A5 y 2 − A6 .

σx x =

(7.34)

σ yy

(7.35)

σx y

(7.36)

The boundary conditions are



b −b

σ yx = 0 ; y = ±b

(7.37)

σ yy = 0 ; y = ±b

(7.38)

yσx x dy = 0 ; x = ±a ,

(7.39)

where we note that weak boundary conditions are imposed on the ends x = ±a. Since the solution is odd in y, only the moment and shear force conditions are non-trivial and the latter need not be explicitly imposed, since they will be satisfied by global equilibrium (as in the example in §5.2.2). Conditions (7.37, 7.38) have to be satisfied for all x and hence the corresponding coefficients of all powers of x must be zero. It follows immediately that A1 = 0 and the remaining conditions can be written

7.4 Rotational Acceleration

111

1 − ρΩ˙ + 6A2 b2 + 6A4 = 0 2 1 − ρΩ˙ − 9A2 b2 − 3A4 = 0 4 1 ˙ 2 − 5A3 b4 − 3A5 b2 − A6 = 0 . ρΩb 4 We get one additional condition from the requirement that φ (equation (7.33)) be biharmonic 72 A2 + 120 A3 = 0 and another from the boundary condition (7.39), which with (7.34) yields 1 ρΩ˙ + 2 A2 a 2 + 4 A3 b2 + 2 A5 = 0 . 6 The solution of these equations is routine, giving the stress function φ= −

ρΩ˙ (5x 3 y 3 − 3x y 5 − 10b2 x 3 y+11b2 x y 3 −5a 2 x y 3 +15a 2 b2 x y−33b4 x y) . 60b2

The complete stress field, including the particular solution terms, is σx x

= ρΩ˙ x y



y2 3 (a 2 − x 2 ) + − b2 5 2b2

˙ 2 − y2) σx y = ρΩ(b



 ; σ yy

ρΩ˙ x y = 2

y 2 + a 2 − 3x 2 11 − 4b2 20

  y2 1− 2 b

 ,

from (7.34–7.36). Finally, we can determine the forces F on the ends by integrating the shear traction, σx y over either end as  F =−

b −b

σx y (a)dy =

2 2 ˙ ρΩb(a + b2 ) . 3

(7.40)

Maple and Mathematica solutions of this problem are given in the files ‘S742’. As in §7.4.1, the relation between the applied loading and the angular acceleration has been obtained without recourse to the equations of rigid-body dynamics. However, we note that the moment of inertia of the rectangular bar for rotation about the origin is I = 4ρab(a 2 +b2 )/3 and the applied moment is M = 2Fa, so application of the equation M = I Ω˙ leads to the same expression for F as that found in (7.40).

112

7 Body Forces

7.4.3 Weak boundary conditions and the equation of motion We saw in §5.2.1 that weak boundary conditions need only be applied at one end of a stationary rectangular bar, since the stress field defined in terms of the Airy stress function must involve tractions that maintain the body in equilibrium. Similarly, in dynamic problems involving prescribed accelerations, an appropriate set of weak boundary conditions can be omitted. For example, we did not have to specify the value of F in the problem of Figure 7.3. Instead, its value was calculated after the stress field had been determined and necessarily proved to be consistent with the equations of rigid-body dynamics. If a body is prevented from moving by a statically determinate support, it is natural to treat the support reactions as the ‘neglected’ weak boundary conditions. Thus, if the body is attached to a rigid support at one boundary, we apply no weak conditions at that boundary, as in §5.2.1. If the body is simply supported at a boundary, we impose the weak boundary condition that there be zero moment applied at the support, leading for example to the conditions stated in footnote 2 on Page 69. Similar considerations apply in a dynamic problem if the body is attached to a support which moves in such a way as to prescribe the acceleration of the body. However, we may also wish to solve problems in which specified non-equilibrated loads are applied to an unsupported body. For example, we may be asked to determine the stresses in the body of Figure 7.3 due to prescribed end loads F. One way to do this would be first to solve a rigid-body dynamics problem to determine the accelerations and then proceed as in §7.4.2. However, in view of the present discussion, a more natural approach would be to include the angular acceleration Ω˙ as an unknown and use equation (7.40) to determine it in terms of F at the end of the solution procedure. More generally, if we have a two-dimensional body subjected to prescribed tractions on all edges, we could assume the most general accelerations (7.9, 7.10) and solve the body force problem, treating a0x , a0y , Ω˙ as if they were known. If strong boundary conditions are imposed on two opposite edges and weak boundary conditions on both the remaining edges, it will then be found that there are three extra conditions which serve to determine the unknown accelerations.

Problems 7.1. Every particle of an elastic body of density ρ experiences a force Cδm F= 2 r directed towards the origin, where C is a constant, r is the distance from the origin and δm is the mass of the particle. Find a body force potential V that satisfies these conditions.

Problems

113

7.2. Verify that the body force distribution px = C y ; p y = −C x is non-conservative, by substituting into equation (7.6). Use the technique of §7.2.4 to attempt to construct a body force potential V for this case. Identify the step at which the procedure breaks down. 7.3. To construct a particular solution for the stress components in plane strain due to a non-conservative body force distribution, it is proposed to start by representing the displacement components in the form 1 ∂ψ 1 ∂ψ ux = ; uy = − ; uz = 0 . 2μ ∂ y 2μ ∂x Use the strain-displacement equations (1.37) and Hooke’s law (1.54) to find expressions for the stress components in terms of ψ. Substitute these results into the equilibrium equations (2.1, 2.2) to find the governing equations for the stress function ψ. What is the condition that must be satisfied by the body force distribution p if these equations are to have a solution? Show that this condition is satisfied if the body force distribution can be written in terms of a potential function W as ∂W ∂W ; py = − . px = ∂y ∂x For the special case px = C y ; p y = −C x , find a particular solution for ψ in the axisymmetric polynomial form ψ = A(x 2 + y 2 )n , where A is a constant and n is an appropriate integer power. Show that this solution can be used to obtain the stress components (7.30–7.32). Suggest ways in which this method might be adapted to give a particular solution for more general nonconservative body force distributions. 7.4. If the elastic displacement u varies in time, there will generally be accelerations a = u¨ and hence body forces p = −ρu¨ , from equation (7.8). Use this result and equation (2.16) to develop the general equation of linear Elastodynamics. Show that this equation is satisfied by a displacement field of the form u x = f (x − c1 t) ; u y = g(x − c2 t) ; u z = 0 , where f, g are any functions and c1 , c2 are two constants that depend on the material properties. Find the values of c1 , c2 and comment on the physical significance of this solution.

114

7 Body Forces

7.5. The beam −b < y < b, 0 < x < L is built-in at the edge x = L and loaded only by its own weight, ρg per unit volume. Find a solution for the stress field, using weak conditions on the end x = 0. 7.6. One wall of a multistory building of height H is approximated as a thin plate −b < x < b, 0 < y < H . During an earthquake, ground motion causes the building to experience a uniform acceleration a in the x-direction. Find the resulting stresses in the wall if the material has density ρ and the edges x = ±b, y = H can be regarded as traction free. 7.7. A tall thin rectangular plate −a < x < a, −b < y < b (b  a) is supported on the vertical edges x = ±a and loaded only by its own weight (density ρ). Find the stresses in the plate using weak boundary conditions on the horizontal edges y = ±b and assuming that the support tractions consist only of uniform shear. 7.8. Figure 7.4(a) shows a triangular cantilever, defined by the boundaries y = 0, y = x tan α and built-in at x = a. It is loaded by its own weight, ρg per unit volume. Find a solution for the complete stress field and compare the maximum tensile bending stress with that predicted by the Mechanics of Materials theory. Would the maximum tensile stress be lower if the alternative configuration of Figure 7.4(b) were used? 7.9. The thin rectangular plate −a < x < a, −b < y < b with a  b rotates about the y-axis at constant angular velocity Ω. All surfaces of the plate are traction free. Find a solution for the stress field, using strong boundary conditions on the long edges y = ±b and weak boundary conditions on the ends x = ±a. 7.10. Solve Problem 7.9 for the case where a  b. In this case you should use strong boundary conditions on x = ±a and weak boundary conditions on y = ±b. 7.11. A thin triangular plate bounded by the lines y = ± x tan α, x = a rotates about the axis x = a at constant angular velocity Ω. The inclined edges of the plate y =

Fig. 7.4 A triangular cantilever

Problems

115

± x tan α are traction free. Find a solution for the stress field, using strong boundary conditions on the inclined edges (weak boundary conditions will then be implied on x = a). 7.12. A thin triangular plate bounded by the lines y = ± x tan α, x = a is initially at rest, but is subjected to tractions at the edge x = a causing an angular acceleration Ω˙ about the perpendicular axis through the point (a, 0). The inclined surfaces y = ± x tan α are traction free. Find a solution for the instantaneous stress field, using the technique of §7.4.2. 7.13. The thin square plate −a < x < a, −a < y < a rotates about the z-axis at constant angular velocity Ω. Use Mathieu’s method (§6.4 and Problem 6.6) to develop a series solution that satisfies all the boundary conditions in the strong sense. Make a contour plot of the maximum principal stress based on truncating the series at m = n = 2, and identify the location and magnitude of the maximum tensile stress. By what percentage does this value differ from the approximation obtained from §7.3.1?

Chapter 8

Problems in Polar Coördinates

Polar coördinates (r, θ) are particularly suited to problems in which the boundaries can be expressed in terms of equations like r = a, θ = α. This includes the stresses in a circular disk or around a circular hole, the curved beam with circular boundaries and the wedge, all of which will be discussed in this and subsequent chapters.

8.1 Expressions for Stress Components We first have to transform the biharmonic equation (4.9) and the expressions for stress components (4.6) into polar coördinates, using the relations r=



x2

x = r cos θ ; y = r sin θ y . + y 2 ; θ = arctan x

(8.1)

The derivation is tedious, but routine. We first note by differentiation that ∂ ∂r ∂ ∂θ ∂ ∂ sin θ ∂ = + = cos θ − ∂x ∂x ∂r ∂x ∂θ ∂r r ∂θ ∂ ∂r ∂ ∂θ ∂ ∂ cos θ ∂ = + = sin θ + . ∂y ∂ y ∂r ∂ y ∂θ ∂r r ∂θ

(8.2) (8.3)

It follows that

Supplementary Information The online version contains supplementary material available at https://doi.org/10.1007/978-3-031-15214-6_8. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 J. R. Barber, Elasticity, Solid Mechanics and Its Applications 172, https://doi.org/10.1007/978-3-031-15214-6_8

117

118

8 Problems in Polar Coördinates

   sin θ ∂ ∂ sin θ ∂ ∂2 ∂ − cos θ − = cos θ ∂x 2 ∂r r ∂θ ∂r r ∂θ  2 2  ∂ 1 ∂ 1 ∂ = cos2 θ 2 + sin2 θ + 2 2 ∂r r ∂r r ∂θ   1 ∂2 1 ∂ − +2 sin θ cos θ 2 r ∂θ r ∂r ∂θ

(8.4)

and by a similar process, we find that   ∂2 1 ∂2 ∂2 1 ∂ 2 2 + 2 2 = sin θ 2 + cos θ ∂ y2 ∂r r ∂r r ∂θ   1 ∂2 1 ∂ − −2 sin θ cos θ 2 r ∂θ r ∂r ∂θ   2 2 ∂ ∂ 1 ∂2 1 ∂ = sin θ cos θ − − ∂x∂ y ∂r 2 r ∂r r 2 ∂θ2     1 ∂ 1 ∂2 − . − cos2 θ − sin2 θ r 2 ∂θ r ∂r ∂θ

(8.5)

(8.6)

Finally, we can determine the expressions for stress components, noting for example that σrr = σx x cos2 θ + σ yy sin2 θ + 2σx y sin θ cos θ

(8.7)

∂ φ ∂ φ ∂ φ + sin2 θ 2 − 2 sin θ cos θ ∂ y2 ∂x ∂x∂ y 1 ∂2φ 1 ∂φ + 2 2 , = r ∂r r ∂θ = cos2 θ

2

2

2

after substituting for the partial derivatives from (8.4–8.6) and simplifying. The remaining stress components, σθθ , σr θ can be obtained by a similar procedure. We find 1 ∂2φ 1 ∂φ ∂2φ + 2 2 ; σθθ = (8.8) σrr = r ∂r r ∂θ ∂r 2   1 ∂φ 1 ∂ 2 φ ∂ 1 ∂φ . (8.9) σr θ = 2 − =− r ∂θ r ∂r ∂θ ∂r r ∂θ If there is a conservative body force p described by a potential V , the stress components are modified to σrr = σr θ

1 ∂2φ 1 ∂φ ∂2φ + 2 2 + V ; σθθ = +V r ∂r r ∂θ ∂r 2   ∂ 1 ∂φ 1 ∂φ 1 ∂ 2 φ − =− , = 2 r ∂θ r ∂r ∂θ ∂r r ∂θ

8.2 Strain Components

119

where pr = −

∂V ∂r

;

pθ = −

1 ∂V . r ∂θ

We also note that the Laplacian operator ∇2 ≡

1 ∂2 ∂2 ∂2 ∂2 1 ∂ + + = + , ∂x 2 ∂ y2 ∂r 2 r ∂r r 2 ∂θ2

(8.10)

from equations (8.4, 8.5) and hence  ∇ 4φ ≡

1 ∂2 ∂2 1 ∂ + + ∂r 2 r ∂r r 2 ∂θ2



∂ 2 φ 1 ∂φ 1 ∂2φ + + ∂r 2 r ∂r r 2 ∂θ2

 .

(8.11)

Notice that in applying the second Laplacian operator in (8.11) it is necessary to differentiate by parts. The two differential operators cannot simply be multiplied together. In Mathematica and Maple, the easiest technique is to define a new function f ≡ ∇2φ =

1 ∂2φ ∂ 2 φ 1 ∂φ + 2 2 + 2 ∂r r ∂r r ∂θ

and then obtain ∇ 4 φ from ∇ 4φ = ∇ 2 f =

1 ∂2 f ∂2 f 1∂f + + . ∂r 2 r ∂r r 2 ∂θ2

8.2 Strain Components A similar technique can be used to obtain the strain-displacement relations in polar coördinates. Writing u x = u r cos θ − u θ sin θ u y = u r sin θ + u θ cos θ and substituting in (8.2), we find ex x =

  ∂u r ∂u θ 1 ∂u r ∂u x uθ = cos2 θ + − − sin θ cos θ ∂x ∂r r ∂r r ∂θ   1 ∂u θ ur + sin2 θ . + r r ∂θ

Using the same method to obtain expressions for ex y , e yy and substituting the results in the strain-transformation relations analogous to (8.7) etc., we obtain the polar coördinate strain-displacement relations

120

err =

8 Problems in Polar Coördinates

∂u r 1 ; er θ = ∂r 2



1 ∂u r ∂u θ uθ + − r ∂θ ∂r r

 ; eθθ =

ur 1 ∂u θ + . r ∂θ r

(8.12)

8.3 Fourier Series Expansion The simplest problems in polar coördinates are those in which there are no θ-boundaries, the most general case being the disk with a central hole illustrated in Figure 8.1 and defined by a π) can only be obtained if the conditions at infinity are also prescribed and satisfied. 12

I. N. Sneddon, Fourier Transforms, McGraw-Hill, New York, 1951. See Chapter 19 and in particular, §19.5. 14 The basic theory of the Mellin transform and its relation to the Fourier transform is explained by I. N. Sneddon, loc. cit.. For applications to elasticity problems for the wedge, see E. Sternberg and W. T. Koiter (1958), The wedge under a concentrated couple: A paradox in the two-dimensional theory of elasticity, ASME Journal of Applied Mechanics, Vol. 25, pp. 575–581; W. J. Harrington and T. W. Ting (1971), Stress boundary-value problems for infinite wedges, Journal of Elasticity, Vol. 1, pp. 65–81. 13

Problems

181

More precisely, we could formulate the problem for the large but finite sector −α < θ < α, 0 0, y = 0, the remaining tractions on y = 0 being zero. Find the complete stress field in the half-plane. Note: This is a problem requiring a special stress function with a logarithmic multiplier (see §11.1.4 above).

Fig. 11.9 The half-plane with shear tractions.

11.2. The half-plane y < 0 is subjected to a uniform normal pressure σ yy = −S on the half-line, x > 0, y = 0, the remaining tractions on y = 0 being zero. Find the complete stress field in the half-plane. 11.3. Find the displacements u r , u θ due to the stress function (11.8) and hence show that the strains are everywhere bounded, but the rotation ω is logarithmically

15

A problem of this kind was considered by G. Tsamasphyros and P. S. Theocaris (1979), On the solution of the sector problem, Journal of Elasticity, Vol. 9, pp. 271–281. 16 This requires that the eigenfunction series is complete for this problem. The proof is given by R. D. Gregory (1979), Green’s functions, bi-linear forms and completeness of the eigenfunctions for the elastostatic strip and wedge, Journal of Elasticity, Vol. 9, pp. 283–309.

182

11 Wedge Problems

unbounded at the corner. You can assume without proof that the expression for ω in polar coordinates is 1 ω= 2



∂u r 1 ∂u θ − ∂r r ∂θ



 cos(2θ) −

∂u θ 1 ∂u r + ∂r r ∂θ



 sin(2θ) .

11.4. The wedge −α < θ < α is loaded by a concentrated moment M0 at the apex, the plane edges being traction free. Use dimensional arguments to show that the stress components must all have the separated-variable form σ=

f (θ) . r2

Use this result to choose a suitable stress function and hence find the complete stress field in the wedge. 11.5. Show that φ = Ar 2 θ can be used as a stress function and determine the tractions which it implies on the boundaries of the region −π/2 < θ < π/2. Hence show that the stress function appropriate to the loading of Figure 11.10 is φ=−

F 2 (r θ1 + r22 θ2 ) 4πa 1

where r1 , r2 , θ1 , θ2 are defined in the Figure. Determine the principal stresses at the point B.

θ

θ

Fig. 11.10 Uniform loading over a discrete region.

11.6. A wedge-shaped concrete dam is subjected to a hydrostatic pressure ρgh varying with depth h on the vertical face θ = 0 as shown in Figure 11.11, the other

Problems

183

face θ = α being traction free. The dam is also loaded by self-weight, the density of concrete being ρc = 2.3ρ. Find the minimum wedge angle α if there is to be no tensile stress in the dam.

ρ ρ Fig. 11.11 The wedgeshaped dam.

α ρ

11.7. Figure 11.12 shows a 45o triangular plate ABC built in at BC and loaded by a uniform pressure p0 on the upper edge AB. The inclined edge AC is traction free. Find the stresses in the plate, using weak boundary conditions on the edge BC (which do not need to be explicitly enforced). Hence compare the maximum tensile stress with the prediction of the elementary bending theory.

Fig. 11.12 A triangular plate

184

11 Wedge Problems

11.8. The triangular body −α < θ < α, x =r cos θ < a is supported at x = a and loaded by uniform antisymmetric shear tractions σθr (r, ±α) = S. Find the Cartesian stress components σx x , σ yy , σx y in the body using weak boundary conditions on x = a and strong conditions on the inclined surfaces. What force and/or moment resultants are transmitted across the surface x = a? 11.9. The wedge −α < θ < α is bonded to a rigid body on both edges θ = ±α. Use the eigenfunction expansion of §11.2.2 to determine the characteristic equations that must be satisfied by the exponent λ in the stress function (11.18) for symmetric and antisymmetric stress fields. Show that these equations reduce to (11.29, 11.30) if ν = 0.5 and plane strain conditions are assumed. 11.10. Find the equation that must be satisfied by λ if the stress function (11.18) is to define a non-trivial solution of the problem of the half-plane 0 < θ < π, traction free on θ = 0 and in frictionless contact with a rigid plane surface at θ = π. Find the lowest value of λ that satisfies this equation and obtain explicit expressions for the form of the corresponding singular stress field near the corner. This solution is of importance in connection with the frictionless indentation of a smooth elastic body by a rigid body with a sharp corner. 11.11. Two large bodies of similar materials with smooth, continuously differentiable curved surfaces make frictionless contact. Use the asymptotic method of §11.2 to examine the asymptotic stress fields near the edge of the resulting contact area. The solution of this problem must also satisfy two inequalities: that the contact tractions in the contact region must be compressive and that the normal displacements in the non-contact region must cause a non-negative gap. Show that one or other of these conditions is violated, whatever sign is taken for the multiplier on the first eigenfunction. What conclusion do you draw from this result? 11.12. Find the equation that must be satisfied by λ if the stress function (11.18) is to define a non-trivial solution of the problem of the wedge 0 < θ < π/2, traction free on θ = 0 and bonded to a rigid plane surface at θ = π/2. Do not attempt to solve the equation.

Chapter 12

Plane Contact Problems

In the previous chapter, we considered problems in which the infinite wedge was loaded on its faces or solely by tractions on the infinite boundary. A related problem of considerable practical importance concerns the wedge with traction-free faces, loaded by a concentrated force F at the vertex, as shown in Figure 12.1.1

β α

θ

Fig. 12.1 The wedge loaded by a force at the vertex.

12.1 Self-Similarity An important characteristic of this problem is that there is no inherent length scale. An enlarged photograph of the problem would look the same as the original. The 1

For a more detailed discussion of elastic contact problems, see K. L. Johnson, Contact Mechanics, Cambridge University Press, 1985, G. M. L. Gladwell, Contact Problems in the Classical Theory of Elasticity, Sijthoff and Noordhoff, Alphen aan den Rijn, 1980, and J. R. Barber Contact Mechanics, Springer, Cham, 2018. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 J. R. Barber, Elasticity, Solid Mechanics and Its Applications 172, https://doi.org/10.1007/978-3-031-15214-6_12

185

186

12 Plane Contact Problems

solution must therefore share this characteristic and hence, for example, contours of the stress function φ must have the same geometric shape at all distances from the vertex. Problems of this type — in which the solution can be mapped into itself after a change of length scale — are described as self-similar. An immediate consequence of the self-similarity is that all the stress components must be capable of expression in the separated-variable form σ = f (r )g(θ) . Furthermore, since the tractions on the line r = a must balance a constant force F for all a, we can deduce that f (r ) =r −1 , because the area available for transmitting the force increases linearly with radius2 .

12.2 The Flamant Solution Choosing those terms in Table 8.1 that give stresses proportional to r −1 , we obtain φ = C1r θ sin θ + C2 r θ cos θ + C3r ln(r ) cos θ + C4 r ln(r ) sin θ ,

(12.1)

for which the stress components are σrr = r −1 (2C1 cos θ − 2C2 sin θ + C3 cos θ + C4 sin θ) σr θ = r −1 (C3 sin θ − C4 cos θ) σθθ = r −1 (C3 cos θ + C4 sin θ) . For the wedge faces to be traction free, we require σθr = σθθ = 0 ; θ = α, β , which can be satisfied by taking C3 , C4 = 0. To determine the remaining constants C1 , C2 , we consider the equilibrium of the region 0 0), we have 2μu r = 2μu x = 0  Fy  2μu θ = 2μu y = − (κ + 1) ln(r ) − 1 , 2π

(12.7) (12.8)

whilst on θ = −π, (y = 0, x < 0), we have Fy (κ − 1) 2  Fy  (κ + 1) ln(r ) + 1 . 2μu θ = −2μu y = 2π 2μu r = −2μu x =

(12.9) (12.10)

It is convenient to impose symmetry on the solution by superposing a rigid-body displacement Fy (κ − 1) Fy ; 2μu y = , 2μu x = 4 2π after which equations (12.7–12.10) can be summarised in the form Fy (κ − 1)sgn(x) 8μ Fy (κ + 1) ln |x| , u y (x, 0) = − 4πμ

u x (x, 0) =

(12.11) (12.12)

12.3 The Half-Plane

189

where the signum function sgn(x) is defined to be +1 for x > 0 and −1 for x < 0. Note incidentally that r = |x| on y = 0.

12.3.2 The tangential force Fx The tangential force Fx produces the stress field σrr = −

2Fx cos θ , πr

which is compressive ahead of the force (θ = 0) and tensile behind it (θ = −π) as we might expect. The corresponding displacement field is found from Table 9.1 as before and after superposing an appropriate rigid-body displacement as in §12.3.1, the surface displacements can be written u x (x, 0) = −

Fx (κ + 1) ln |x| Fx (κ − 1)sgn(x) ; u y (x, 0) = − . 4πμ 8μ

12.3.3 Summary We summarize the results of §§12.3.1, 12.3.2 in the equations Fy (κ − 1)sgn(x) Fx (κ + 1) ln |x| + 4πμ 8μ Fy (κ + 1) ln |x| Fx (κ − 1)sgn(x) − , u y (x, 0) = − 8μ 4πμ

u x (x, 0) = −

see Figure 12.3.

Fig. 12.3 Surface displacements due to a surface force.

(12.13) (12.14)

190

12 Plane Contact Problems

12.4 Distributed Normal Tractions Now suppose that the surface of the half-plane is subjected to a distributed normal load p(ξ) per unit length as shown in Figure 12.4. The stress and displacement field can be found by superposition, using the Flamant solution as a Green’s function — i.e. treating the distributed load as the limit of a set of concentrated loads of magnitude p(ξ)δξ. Of particular interest is the distortion of the surface, defined by the normal displacement u y . At the point P(x, 0), this is given by (κ + 1) u y (x, 0) = − 4πμ

 A

p(ξ) ln |x − ξ|dξ ,

(12.15)

from equation (12.12), where A is the region over which the load acts.

ξ δξ

ξ

δξ

Fig. 12.4 Half-plane subjected to a distributed traction.

The displacement defined by equation (12.15) is logarithmically unbounded as x tends to infinity. In general, if a finite region of the surface of the half-plane is subjected to a non-self-equilibrated system of loads, the stress and displacement fields at a distance r  A will approximate those due to the force resultants applied as concentrated forces — i.e. to the expressions of equations (12.4–12.6) for normal loading. In other words, the stresses will decay with 1/r and the displacements (being integrals of the strains) will vary logarithmically. This means that the half-plane solution can be used for the stresses in a finite body loaded on a region of the surface which is small compared with the linear dimensions of the body, since 1/r is arbitrarily small for sufficiently large r . However, the logarithm is not bounded at infinity and hence the rigid-body displacement of the loaded region with respect to distant parts of the body cannot be found without a more exact treatment taking into account the finite dimensions of the body. For this reason, contact problems for the half-plane are often formulated in terms of the displacement gradient

12.5 Frictionless Contact Problems

191

du y (κ + 1) (x, 0) = − dx 4πμ

 A

p(ξ)dξ , (x − ξ)

(12.16)

thereby avoiding questions of rigid-body translation. When the point x = ξ lies within A, the integral in (12.16) is interpreted as a Cauchy principal value — for example  a

b

p(ξ)dξ = lim (x − ξ) →0

 a

x−

 +

b



x+

p(ξ)dξ ; a 0 for all t)12 . The related solution for two spheres in contact (also due to Mindlin) has been verified experimentally by observing the damaged regions produced by cyclic microslip between tangentially loaded spheres13 . A significant generalization of these results has recently been discovered independently by Jäger14 and Ciavarella15 , who showed that the frictional traction distribution satisfying both equality and inequality conditions for any plane frictional contact problem (not necessarily Hertzian) will consist of a superposition of the limiting friction distribution and an opposing distribution equal to the coefficient of friction multiplied by the normal contact pressure distribution at some smaller value of the normal load. Thus, as the tangential force is increased at constant normal force, the stick zone shrinks, passing monotonically through the same sequence of areas as the normal contact area passed through during the normal loading process. These results can be used to predict the size of the slip zone in conditions of fretting fatigue16 . One consequence of this result is that wear in the sliding regions due to an oscillating tangential load will not change the extent of the adhesive region, so that in the limit the contact is pure adhesive and a singularity develops in the normal traction at the edge of this region17 .

12.8.2 Steady rolling: Carter’s solution A final example of considerable practical importance is that in which two cylinders roll over each other whilst transmitting a constant tangential force, T . If we assume that the rolling velocity is V , the solution will tend to a steady-state which is invariant 12

The effect of non-monotonic loading in the related problem of two contacting spheres was considered by R. D. Mindlin and H. Deresiewicz (1953), Elastic spheres in contact under varying oblique forces, ASME Journal of Applied Mechanics, Vol. 21, pp. 327–344. The history-dependence of the friction law leads to quite complex arrangements of slip and stick zones and consequent variation in the load-compliance relation. These results also find application in the analysis of oblique impact, where neither normal nor tangential loading is monotonic (see N. Maw, J. R. Barber and J. N. Fawcett (1976), The oblique impact of elastic spheres, Wear, Vol. 38, pp. 101–114). 13 K. L. Johnson (1961), Energy dissipation at spherical surfaces in contact transmitting oscillating forces, Journal of Mechanical Engineering Science, Vol. 3, pp. 362–368. 14 J. Jäger (1997), Half-planes without coupling under contact loading. Archive of Applied Mechanics Vol. 67 , pp. 247–259. 15 M. Ciavarella (1998), The generalized Cattaneo partial slip plane contact problem. I-Theory, II-Examples. International Journal of Solids and Structures. Vol. 35, pp. 2349–2378. 16 D. A. Hills and D. Nowell, Mechanics of Fretting Fatigue. Kluwer, Dordrecht, 1994, M. P. Szolwinski and T. N. Farris (1996), Mechanics of fretting fatigue crack formation. Wear Vol. 198, pp. 93–107. 17 M. Ciavarella and D. A. Hills (1999), Some observations on the oscillating tangential forces and wear in general plane contacts, European Journal of Mechanics A–Solids. Vol. 18, pp. 491–497.

12.8 Combined Normal and Tangential Loading

207

with respect to the moving coördinate system ξ = x − Vt . In this system, we can write the ‘stick’ condition (12.49) as d (1) d  (2) ˙ ˙ u x (x − V t) − u (1) (u − u (2) h(x, t) = x (x − V t) + C = V x )+C =0 , dt dξ x (12.54) where C˙ is an arbitrary but constant rigid-body slip (or creep) velocity. Similarly, the slip condition (12.50) becomes   d (1) (2) u − ux + C˙ . px (ξ) = − f p y (ξ) sgn V dξ x 

(12.55)

At first sight, we might think that the Cattaneo traction distribution (12.53) satisfies this condition, since it gives d (1) (u − u (2) x )=0 dξ x in the central stick zone and C˙ can be chosen arbitrarily. However, if we substitute the resulting displacements into the slip condition (12.55), we find a sign error in the leading slip zone. This can be explained as follows: In the Cattaneo problem, as T is ˙ = u˙ x2 − u˙ x1 + C˙ > 0) occurs in both slip zones and increased, positive slip (i.e. h(x) the magnitude of h(x) increases from zero at the stick-slip boundary to a maximum d (2) (u (1) at x = ±a. It follows that dξ x −u x ) is negative in the right slip zone and positive in the left slip zone. Thus, if this solution is used for the steady rolling problem, a violation of (12.55) will occur in the right zone if V is positive, and in the left zone if V is negative. In each case there is a violation in the leading slip zone — i.e. in that zone next to the edge where contact is being established. Now in frictional problems, when we make an assumption that a given region slips and then find that it leads to a sign violation, it is usually an indication that we made the wrong assumption and that the region in question should be in a state of stick. Thus, in the rolling problem, there is no leading slip zone18 . Carter19 has shown that the same kind of superposition can be used for the rolling problem as for Cattaneo’s problem, except that the corrective traction is displaced to a zone adjoining the leading edge. A corrective traction

Note that this applies to the uncoupled problem (β = 0) only. With dissimilar materials, there is generally a leading slip zone and there can also be an additional slip zone contained within the stick zone. This problem is treated by R. H. Bentall and K. L. Johnson (1967), Slip in the rolling contact of dissimilar rollers, International Journal of Mechanical Sciences, Vol. 9, pp. 389–404. 19 F. W. Carter (1926), On the action of a locomotive driving wheel, Proceedings of the Royal Society of London, Vol. A112, pp. 151–157. 18

208

12 Plane Contact Problems

2fF px∗ (ξ) = πa 2



a−c 2

2

  a+c 2 2f F − ξ− = (a − ξ)(ξ − c) 2 πa 2

produces a displacement distribution    a+c f FA d (1) (2) ξ− u − ux = ; c 0 by a force F. A tangential force T is then applied to the punch. If Coulomb friction conditions apply at the interface with coefficient f and Dundurs constant β = 0, show that no microslip will occur until T reaches the value f F at which point gross slip starts. 12.8. Express the stress function φ=

Fy r θ cos θ π

in Cartesian coördinates and hence find the stress components σx x , σx y , σ yy due to the point force Fy in Figure 12.2. Use this result and the integration procedure of §12.4 (Figure 12.4) to determine the stress field in the half-plane y < 0 due to the Hertzian traction distribution of equation (12.35). Make a contour plot of the von Mises stress σ E of equation (1.23) and identify the maximum value and its location. 12.9. From §12.5.2 and equation (12.30), it follows that the contact traction distribution

Problems

213

1 p(x, a) = √ ; |x| < a 2 a − x2 =0 ; |x| > a produces zero surface slope du y /d x in |x| < a. Use equation (12.16) to determine the corresponding value of slope outside the contact area (|x| > a) and hence construct the discontinuous function du y , u(x, a) ≡ dx such that p(x, a) produces u(x, a). Linear superposition then shows that the more general traction distribution 

b

p(x) =

g(a) p(x, a)da 0

produces the surface slope du y = dx



b

g(a)u(x, a)da ,

0

where g(a) is any function of a. In effect this is a superposition of a range of ‘flat punch’ traction distributions over different width strips up to a maximum semi-width of b, so the traction will still be zero for |x| > b. Use this representation to solve the problem of the indentation by a wedge of semi-angle π/2−α (α 1), for which du 0 = −|α| ; |x| < b , dx where b is the semi-width of the contact area. In particular, find the contact traction distribution and the relation between b and the applied force F. Hint: You will find that the boundary condition leads to an Abel integral equation, whose solution is given in Table 32.2 in Chapter 32. 12.10. Modify the method outlined in Problem 12.9 to solve the problem of the indentation of a half-plane by an unsymmetrical wedge, for which du 0 = −α ; 0 < x < c dx = β ; −d < x < 0 , where the unsymmetrical contact area extends from −d to c. In particular, find the contact traction distribution and both c and d as functions of the applied force F. The modification involves moving the origin to destroy the symmetry once the function u(x, a) has been determined. In other words, instead of superposing a set of

214

12 Plane Contact Problems

symmetrically disposed ‘flat punch’ distributions, superpose a similar set arranged so that the common point is displaced from the origin.

Chapter 13

Forces, Dislocations and Cracks

In this chapter, we shall discuss the applications of two solutions which are singular at an interior point of a body, and which can be combined to give the stress field due to a concentrated force (the Kelvin solution) and a dislocation. Both solutions involve a singularity in stress with exponent −1 and are therefore inadmissable according to the criterion of §11.2.1. However, like the Flamant solution considered in Chapter 12, they can be used as Green’s functions to describe distributions, resulting in convolution integrals in which the singularity is integrated out. The Kelvin solution is also useful for describing the far field (i.e. the field a long way away from the loaded region) due to a force distributed over a small region.

13.1 The Kelvin Solution We consider the problem in which a concentrated force F acts in the x-direction at the origin in an infinite body. This is not a perturbation problem like the stress field due to a hole in an otherwise uniform stress field, since the force has a non-zero resultant. Thus, no matter how far distant we make the boundary of the body, there will have to be some traction to oppose the force. In fact, self-similarity arguments like those used in §§12.1, 12.2 show that the stress field must decay with r −1 . The Flamant solution has this behaviour and it corresponds to a concentrated force (see §12.2), but it cannot be used at an interior point in the body, since the

Supplementary Information The online version contains supplementary material available at https://doi.org/10.1007/978-3-031-15214-6_13.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 J. R. Barber, Elasticity, Solid Mechanics and Its Applications 172, https://doi.org/10.1007/978-3-031-15214-6_13

215

216

13 Forces, Dislocations and Cracks

corresponding displacements [equations (12.5, 12.6)] are multivalued1 . However, we can construct a solution with the same character and with single-valued displacements from the more general stress function (12.1) by choosing the coefficients in such a way that the multivalued terms cancel. In view of the symmetry of the problem about θ = 0, we restrict attention to the symmetric terms (13.1) φ = C1r θ sin θ + C3r ln(r ) cos θ of (12.1), for which the stress components are σrr =

2C1 cos θ C3 cos θ C3 sin θ + ; σr θ = ; r r r

σθθ =

C3 cos θ , r

(13.2)

from Table 8.1, and the displacement components are  C1  (κ − 1)θ sin θ − cos θ + (κ + 1) ln(r ) cos θ 2  C3  (κ + 1)θ sin θ − cos θ + (κ − 1) ln(r ) cos θ + 2  C1  (κ − 1)θ cos θ − sin θ − (κ + 1) ln(r ) sin θ 2μu θ = 2  C3  (κ + 1)θ cos θ − sin θ − (κ − 1) ln(r ) sin θ , + 2 2μu r =

(13.3)

(13.4)

from Table 9.1. Suppose we make an imaginary cut in the plane at θ = 0, 2π and define a principal value of θ such that 0 ≤ θ < 2π. This makes θ discontinuous at θ = 2π, but the trigonometric functions of course remain continuous. The only potential difficulty is associated with the expressions θ sin θ, θ cos θ. Now θ sin θ = 0 at θ = 2π and hence this expression has the same value at the two sides of the cut and is continuous. We can therefore make the whole displacement field continuous by choosing C1 , C3 such that the terms θ cos θ in u θ cancel — i.e. by setting2 C1 (κ − 1) + C3 (κ + 1) = 0 , (13.5) which for plane stress is equivalent to (1 − ν)C1 + 2C3 = 0 .

1

(13.6)

This was not a problem for the surface loading problem, since the wedge of Figure 12.1 only occupies a part of the θ-domain and hence a suitable principal value of θ can be chosen to be both single-valued and continuous. 2 Notice that this choice also has the effect of cancelling the θ sin θ terms in (13.3), so that the complete displacement field, like the stress field, depends on θ only through sine and cosine terms.

13.1 The Kelvin Solution

217

This leaves us with one degree of freedom (one free constant) to satisfy the condition that the force at the origin is equal to F. Considering the equilibrium of a small circle of radius r surrounding the origin, we have 

2π

F+

(σrr cos θ − σr θ sin θ) r dθ = 0 .

0

We now substitute for the stress components from equations (13.2), obtaining C1 = −

F , 2π

(13.7)

after which we recover the constant C3 from equation (13.6) as C3 =

(1 − ν)F . 4π

Finally, we substitute these constants back into (13.2) to obtain σrr = −

(3 + ν)F cos θ (1 − ν)F sin θ (1 − ν)F cos θ ; σr θ = ; σθθ = . 4πr 4πr 4πr

These are the stress components for Kelvin’s problem, where the force acts in the xdirection. The corresponding results for a force in the y-direction can be obtained in the same way, using the antisymmetric terms in (12.1). Alternatively, we can simply rotate the axis system in the above solution by redefining θ → (θ−π/2).

13.1.1 Body force problems Kelvin’s problem is a special case of a body force problem — that in which the body force is a delta function at the origin. The solution can also be used to solve more general body force problems by convolution. We consider the body force px δxδ y acting on the element δxδ y as a concentrated point force and use the above solution to determine its effect on the stress components at an arbitrary point. Treating the component p y in the same way and summing over all the elements of the body then gives a double integral representation of the stress field. This method is not restricted to the infinite body, since we only seek a particular solution of the body force problem. We can therefore use the convolution method to develop a solution for the stresses in an infinite body with the prescribed body force distribution, after which we ‘cut out’ the shape of the real body and correct the boundary conditions as required, using an appropriate homogeneous solution (i.e. a solution without body forces). Any body force distribution can be treated this way — the method is not restricted to conservative vector fields. It is generally more algebraically tedious than the methods developed in Chapter 7, but it lends itself naturally to numerical implementation.

218

13 Forces, Dislocations and Cracks

For example, it can be used to extend the boundary integral method to body force problems.

13.2 Dislocations The term dislocation has related, but slightly different meanings in Elasticity and in Materials Science. In Elasticity, the material is assumed to be a continuum — i.e. to be infinitely divisible. Suppose we take an infinite continuous body and make a cut along the half-plane x > 0, y = 0. We next apply equal and opposite tractions to the two surfaces of the cut such as to open up a gap of constant thickness, as illustrated in Figure 13.1. We then slip a thin slice of the same material into the space to keep the surfaces apart and weld the system up, leaving a new continuous body which will now be in a state of residual stress.

Fig. 13.1 The climb dislocation solution.

The resulting stress field is referred to as the climb dislocation solution. It can be obtained from the stress function of equation (13.1) by requiring that there be no net force at the origin, and hence that C1 = 0 [see equation (13.7)]. We therefore have φ = C3r ln(r ) cos θ .

(13.8)

The strength of the dislocation can be defined in terms of the discontinuity in the displacement u θ on θ = 0, 2π, which is also the thickness of the slice of extra material which must be inserted to restore continuity. This thickness is δ = u θ (0) − u θ (2π) = −

π(κ + 1)C3 . 2μ

13.2 Dislocations

219

Thus, we can define a climb dislocation of strength B y as one which opens a gap δ = B y , corresponding to C3 = −

2μB y π(κ + 1)

or

φ=−

2μB y r ln(r ) cos θ , π(κ + 1)

(13.9)

where μ(1 + ν) E 2μ = = (κ + 1) 2 4 μ E = = 2(1 − ν) 4(1 − ν 2 )

(plane stress) (plane strain),

(13.10)

using equation (3.11). The stress field due to the climb dislocation is σrr = σθθ = −

2μB y cos θ 2μB y sin θ ; σr θ = − . π(κ + 1) π(κ + 1)r

(13.11)

We also record the stress components at y = 0, — i.e. θ = 0, π — in rectangular coördinates, which are σx x = σ yy = −

2μB y ; σ yx = 0 . π(κ + 1)x

(13.12)

This solution is called a climb dislocation, because it opens a gap on the cut at θ = 0, 2π. A corresponding solution can be obtained from the stress function φ=

2μBx r ln(r ) sin θ π(κ + 1)

(13.13)

which is discontinuous in the displacement component u r — i.e. for which the two surfaces of the cut experience a relative tangential displacement but do not separate. This is called a glide dislocation. The stress field due to a glide dislocation is given by 2μBx sin θ 2μBx cos θ ; σr θ = − , (13.14) σrr = σθθ = π(κ + 1)r π(κ + 1)r and on the surface y = 0, these reduce to σx x = σ yy = 0 ; σ yx = −

2μBx , π(κ + 1)x

where Bx = u r (0) − u r (2π) is the strength of the dislocation. The solutions (13.11) and (13.14) actually differ only in orientation. The climb dislocation solution defines a glide dislocation if we choose to make the cut along

220

13 Forces, Dislocations and Cracks

the line θ = −3π/2, π/2 (i.e. along the y-axis instead of the x-axis3 ). The dislocation strengths Bx , B y can be regarded as the components of a vector B = {Bx , B y }, known as the Burgers vector.

13.2.1 Dislocations in Materials Science Real materials have a discrete atomic or molecular structure. However, we can follow a similar procedure by imagining cleaving the solid between two sheets of molecules up to the line x = y = 0 and inserting one extra layer of molecules. When the system is released, there will be some motion of the molecules, mostly concentrated at the end of the added layer, resulting in an imperfection in the regular molecular array. This is what is meant by a dislocation in Materials Science. Considerable insight into the role of dislocations in material behaviour was gained by the work of Bragg4 with bubble models. Bragg devised a method of generating a two-dimensional collection of identical size bubbles. Attractive forces between the bubbles ensured that they adopted a regular array wherever possible, but dislocations are identifiable as can be seen in Figure 13.25 .

Fig. 13.2 The bubble model — a dislocation. The extra layer is most easily seen by looking along a line inclined at ±30◦ to the vertical.

When forces are applied to the edges of the bubble assembly, the dislocations are found to move in such a way as to permit the boundaries to move. This dislocation motion is believed to be the principal mechanism of plastic deformation in ductile materials. Larger bubble assemblies exhibit discrete regions in which the arrays are 3

In fact, the cut can be made along any (not necessarily straight) line from the origin to infinity. The solution of equation (13.11) will then exhibit a discontinuity in u y of magnitude B y at all points along the cut, as long as the principal value of θ is appropriately defined. 4 L. Bragg and J. F. Nye (1947), A dynamical model of a crystal structure, Proceedings of the Royal Society of London, Vol. A190, pp. 474–481. 5 Figures 13.2 and 13.3 are reproduced by kind permission of the Royal Society of London.

13.2 Dislocations

221

differently aligned as in Figure 13.3. These are analogous with grains in multigranular materials. When the structure is deformed, the dislocations typically move until they reach a grain boundary, but the misalignment prohibits further motion and the stiffness of the assembly increases. This pile-up of dislocations at grain boundaries is responsible for work-hardening in ductile materials. Also, the accumulated dislocations coalesce into larger disturbances in the crystal structure such as voids or cracks, which function as initiation points for failure by fatigue or fracture.

Fig. 13.3 The bubble model — grain boundaries.

13.2.2 Similarities and differences It is tempting to deduce that the elastic solution of §13.2 describes the stresses due to the molecular structure dislocation, if we multiply by a constant defining the thickness of a single layer of molecules. However, the concept of stress is rather vague over dimensions comparable with interatomic distances. In fact, this is preëminently a case where the apparent singularity of the mathematical solution is moderated in reality by the discrete structure of the material. Notice also that the continuum dislocation of §13.2 can have any strength, corresponding to the fact that Bx , B y are arbitrary real constants. By contrast, the thickness of the inserted layer is restricted to one layer of molecules in the discrete theory. This thickness is sufficiently small to ensure that the ‘stresses’ due to a material dislocation are very small in comparison with typical engineering magnitudes, except in the immediate vicinity of the defect. Furthermore, although a stress-relieved metal-

222

13 Forces, Dislocations and Cracks

lic component will generally contain numerous dislocations, they will be randomly oriented, leaving the components essentially stress-free on the macroscopic scale. However, if the component is plastically deformed, the dislocation motion will lead to a more systematic structure, which can result in residual stress. Indeed, one way to represent the stress field due to plastic deformation is as a distribution of mathematical dislocations. For example, the motion of a single dislocation can be represented by introducing a dislocation pair comprising a negative dislocation at the original location and an equal positive dislocation at the final location. This implies a closure condition as in §§13.2.3, 13.2.4 below.

13.2.3 Dislocations as Green’s functions In the Theory of Elasticity, the principal use of the dislocation solution is as a Green’s function to represent localized processes. In this context, it is more natural to think of a dislocation density B(x, y) per unit area6 , so that the elemental area d xd y contains a dislocation of Burger’s vector B(x, y)d xd y. Suppose we place a distribution of dislocations in some interior domain Ω of the body, which is completely surrounded by elastic material. The resulting stress field obtained by integration will satisfy the conditions of equilibrium everywhere (since the dislocation solution involves no force) and will satisfy the compatibility condition everywhere except in Ω. There is of course the possibility that the displacement may be multiple-valued outside Ω, but we can prevent this by enforcing the two closure conditions   Bx (x, y)d xd y = 0 ; B y (x, y)d xd y = 0 , (13.15) Ω

Ω

which state that the total strength of the dislocations in Ω is zero7 .

13.2.4 Stress concentrations A special case of some importance is that in which the enclosed domain Ω represents a hole which perturbs the stress field in an elastic body. It may seem strange to place dislocations in a region which is strictly not a part of the body. However, we might start with an infinite body with no hole, place

6

P. M. Blomerus, D. A. Hills and P. A. Kelly (1999), The distributed dislocation method applied to the analysis of elastoplastic strain concentrations, Journal of the Mathematics and Physics of Solids, Vol.47, pp. 1007–1026. 7 This does not mean that the stress field is null, since the various self-cancelling dislocations have different locations.

13.3 Crack Problems

223

dislocations in Ω generating a stress field and then make a cut along the boundary of Ω producing the body with a hole. The stress field will be unchanged by the cut provided we place tractions on its boundary equal to those which were transmitted across the same surface in the original continuous body. In particular, if we choose the dislocation distribution so as to make the boundaries of Ω traction free, we can cut out the hole without changing the stress distribution, which is therefore the solution of the original problem for the body with a hole. The general idea of developing perturbation solutions by placing singularities in a region where the governing equations (here the compatibility condition) are not required to be enforced is well-known in many branches of Applied Mechanics. For example, the solution for the flow of a fluid around a rigid body can be developed in many cases by placing an appropriate distribution of sources and sinks in the region occupied by the body, the distribution being chosen so as to make the velocity component normal to the body surface be everywhere zero. The closure conditions (13.15) ensure that acceptable dislocation distributions can be represented in terms of dislocation pairs — i.e. as matched pairs of dislocations of equal magnitude and opposite direction. It follows that acceptable distributions can be represented as distributions of dislocation derivatives, since, for example, a dislocation at P and an equal negative dislocation at Q is equivalent to a uniform distribution of derivatives on the straight line joining P and Q 8 . Now the stress components in the dislocation solution [equation ( 13.11)] decay with r −1 and hence those in the dislocation derivative solution will decay with r −2 . It follows that the dominant term in the perturbation (or corrective) solution due to a hole will decay at large r with r −2 . We see this in the particular case of the circular hole in a uniform stress field (§§8.3.2, 8.4.1). The same conclusion applies for the perturbation in the stress field due to an inclusion — i.e. a localized region whose properties differ from those of the bulk material (see Chapter 28).

13.3 Crack Problems Crack problems are particularly important in Elasticity because of their relevance to the subject of Fracture Mechanics, which broadly speaking is the study of the stress conditions under which cracks grow. For our purposes, a crack will be defined as the limiting case of a hole whose volume (at least in the unloaded case) has shrunk to zero, so that opposite faces touch. It might also be thought of as an interior surface in the body which is incapable of transmitting tension.

8

To see this, think of the derivative as the limit of a pair of equal and opposite dislocations separated by a distance δS whose magnitude is proportional to 1/δS.

224

13 Forces, Dislocations and Cracks

In practice, cracks will generally have some finite thickness, but if this is sufficiently small, the crack will behave unilaterally with respect to tension and compression — i.e. it will open if we try to transmit tension, but close in compression, transmitting the tractions by means of contact. For this reason, a cracked body will appear stiffer in compression than it does in tension. Also, the crack acts as a stress concentration in tension, but not in compression, so cracks do not generally propagate in compressive stress fields.

13.3.1 Linear Elastic Fracture Mechanics We saw in §11.2.3 that the asymptotic stress field at the tip of a crack has a square-root singularity and we shall find this exemplified in the particular solutions that follow. The simplest and most prevalent theory of brittle fracture — that due to Griffith — states in essence that crack propagation will occur when the scalar multiplier on this singular stress field exceeds a certain critical value. More precisely, Griffith proposed the thesis that a crack would propagate when propagation caused a reduction in the total energy of the system. Crack propagation causes a reduction in strain energy in the body, but also generates new surfaces which have surface energy. Surface energy is related to the force known as surface tension in fluids and follows from the fact that to cleave a solid body along a plane involves doing work against the interatomic forces across the plane. When this criterion is applied to the stress field in particular cases, it turns out that for a small change in crack length, propagation is predicted when the multiplier on the singular term, known as the stress-intensity factor, exceeds a certain critical value, which is a constant for the material known as the fracture toughness. It may seem paradoxical to found a theory of real material behaviour on properties of a singular elastic field, which clearly cannot accurately represent conditions in the precise region where the failure is actually to occur. However, if the material is brittle, non-linear effects will be concentrated in a relatively small process zone surrounding the crack tip. Furthermore, the certainly very complicated conditions in this process zone can only be influenced by the surrounding elastic material and hence the conditions for failure must be expressible in terms of the characteristics of the much simpler surrounding elastic field. As long as the process zone is small compared with the other linear dimensions of the body9 (notably the crack length), it will have only a very localized effect on the surrounding elastic field, which will therefore be adequately characterized by the dominant singular term in the linear elastic solution, whose multiplier (the stress-intensity factor) then determines the conditions for crack propagation.

9

This is often referred to as the small-scale yielding condition.

13.3 Crack Problems

225

It is notable that this argument requires no assumption about or knowledge of the actual mechanism of failure in the process zone and, by the same token, the success of Linear Elastic Fracture Mechanics (LEFM) as a predictor of the strength of brittle components provides no evidence for or against any particular failure theory10 .

13.3.2 Plane crack in a tensile field Two-dimensional crack problems are very conveniently formulated using the methods outlined in §§13.2.3, 13.2.4. Thus, we seek a distribution of dislocations on the plane of the crack (which appears as a line in two-dimensions), which, when superposed on the unperturbed stress field, will make the surfaces of the crack traction free. We shall illustrate the method for the simple case of a plane crack in a tensile stress field.

S

y .

Fig. 13.4 Plane crack in a tensile field.

a

a O

.

x

S

Figure 13.4 shows a plane crack of width 2a occupying the region −a < x < a, y = 0 in a two-dimensional body subjected to uniform tension σ yy = S at its remote boundaries. 10

For more details of the extensive development of the field of Fracture Mechanics, the reader is referred to the many excellent texts on the subject, such as M.F.Kanninen and C.H.Popelar, Advanced Fracture Mechanics, Clarendon Press, Oxford, 1985, H.Leibowitz, ed., Fracture, An Advanced Treatise, 7 Vols., Academic Press, New York, 1971. Stress-intensity factors for a wide range of geometries are tabulated by G.C.Sih, Handbook of Stress Intensity Factors, Institute of Fracture and Solid Mechanics, Lehigh University, Bethlehem, PA, 1973.

226

13 Forces, Dislocations and Cracks

Assuming that the crack opens, the boundary conditions for this problem can be stated in the form σ yx = σ yy = 0 ; −a < x < a, y = 0 σ yy → S ; σx y , σx x → 0 ; r → ∞ . Following the procedure of §8.3.2, we represent the solution as the sum of the stress field in the corresponding body without a crack — here a uniform uniaxial tension σ yy = S — and a corrective solution, for which the boundary conditions are therefore (13.16) σ yx = 0 ; σ yy = −S ; −a < x < a, y = 0 σx x , σx y , σ yy → 0 ; r → ∞ .

(13.17)

Notice that the corrective solution corresponds to the problem of a crack in an otherwise stress-free body opened by uniform compressive normal tractions of magnitude S. The most general stress field due to the crack would involve both climb and glide dislocations, but in view of the symmetry of the problem about the plane y = 0, we conclude that there is no relative tangential motion between the crack faces and hence that the solution can be constructed with a distribution of climb dislocations alone. More precisely, we respresent the solution in terms of a distribution B y (x) of dislocations per unit length in the range −a < x < a, y = 0. We consider first the traction σ yy at the point (x, 0) due to those dislocations between ξ and ξ + δξ on the line y = 0. If δξ is small, they can be considered as a concentrated dislocation of strength B y (ξ)δξ and hence they produce a traction σ yy = −

2μB y (ξ)δξ , π(κ + 1)(x − ξ)

from equation (13.12), since the distance from (ξ, 0) to (x, 0) is (x −ξ). The traction due to the whole distribution of dislocations can therefore be written as the integral  a B y (ξ)dξ 2μ , σ yy = − π(κ + 1) −a (x − ξ) and the boundary condition (13.16) leads to the following Cauchy singular integral equation for B y (ξ) 

a

−a

B y (ξ)dξ π(κ + 1)S = ; −a < x < a . (x − ξ) 2μ

13.3 Crack Problems

227

This is of exactly the same form as equation (12.20) and can be solved in the same way. Writing x = a cos φ ; ξ = a cos θ , we have



π 0

B y (θ) sin θdθ π(κ + 1)S = ; 0 a, y = 0 = S −1 + √ x 2 − a2

σ yy = −

(13.20)

using (13.18) and 3.228.2 of Gradshteyn and Ryzhik11 . Remembering that this is the corrective solution, we add the uniform stress field σ yy = S to obtain the complete stress field, which on the line y = 0 gives σ yy = √

11

S|x| x 2 − a2

; |x| > a, y = 0 .

I.S.Gradshteyn and I.M.Ryzhik, Tables of Integrals, Series and Products, Academic Press, New York, 1980. See also Problem 19.8 for a method of evaluating this integral.

228

13 Forces, Dislocations and Cracks

This tends to the uniform field as it should as x → ∞ and is singular as x → a + . We define12 the mode I stress-intensity factor, K I as  (13.21) K I ≡ lim+ σ yy (x) 2π(x − a) x→a √ Sx 2π(x − a) = lim+ √ x→a x 2 − a2 √ = S πa . (13.22) We can also calculate the crack opening displacement +

−

δ(x) ≡ u y (x, 0 ) − u y (x, 0 ) =



x

−a

B y (ξ)dξ =

(κ + 1)S  2 a − x2 . 2μ

(13.23)

Thus, the crack is opened to the shape of a long narrow ellipse as a result of the tensile field.

13.3.3 Energy release rate If a crack extends as a result of the local stress field, there will generally be a reduction in the total strain energy U of the body. We shall show that the strain-energy release rate, defined as ∂U G=− ∂S where S is the extent of the crack, is a unique function of the stress-intensity factor. We first note that in the immediate vicinity of the crack tip x = a, the stress component σ yy can be approximated by the dominant singular term as KI ; s>0, σ yy (s) = √ 2πs

(13.24)

from equation (13.21) with s = (x −a). A similar approximation for the crack opening displacement δ(s) is K I (κ + 1) S(κ + 1)  2a(−s) = δ(s) = 2μ 2μ



2(−s) , π

using (13.22, 13.23). 12

Notice that the factor 2π in this definition is conventional, but essentially arbitrary. In applying fracture mechanics arguments, it is important to make sure that the results used for the stress-intensity factor (a theoretical or numerical calculation) and the fracture toughness (from experimental data) are based on the same definition of K .

13.3 Crack Problems

229

Fig. 13.5 Geometry of crack extension.

Figure 13.5 shows the configuration of the open crack before and after extension of the crack tip by an infinitesimal distance ΔS. The tractions on the crack plane before crack extension are given by (13.24), where s is measured from the initial position of the crack tip. The crack opening displacement in ΔS after extension is given by  K I (κ + 1) 2(ΔS − s) , (13.25) δ(s) = 2μ π since the point s < ΔS is now a distance ΔS − s to the left of the new crack tip. The reduction in strain energy during crack extension can be found by following a scenario in which the tractions (13.24) are gradually released, allowing work W = −ΔU to be done on them in moving through the displacements (13.25). We obtain 1 W = 2



ΔS 0

K 2 (κ + 1) σ yy (s)δ(s)ds = I 4πμ

and hence G=−

 0



ΔS

K 2 ΔS(κ + 1) (ΔS − s) ds = I s 8μ

∂W K 2 (κ + 1) ∂U = = I . ∂S ∂S 8μ

(13.26)

A similar calculation can be performed for a crack loaded in shear, causing a mode II stress-intensity factor K II . The two deformation modes are orthogonal to each other, so that the energy release rate for a crack loaded in both modes I and II is given by G=

(K I2 + K II2 )(κ + 1) . 8μ

This expression can be written in terms of E, ν, using (13.10). We obtain (K I2 + K II2 ) E (K I2 + K II2 )(1 − ν 2 ) = E

G=

(plane stress)

(13.27)

(plane strain).

(13.28)

230

13 Forces, Dislocations and Cracks

As long as the process zone is small in the sense defined in §13.3.1, a component containing a crack loaded in tension (mode I) will fracture when K I = K Ic

or

G = Gc ,

where K Ic , G c are interrelated material properties, the former being the fracture toughness. The critical energy release rate G c seems to imply a single failure criterion under combined mode I and mode II loading, but this is illusory, since experiments show that the value of G c varies with the mode-mixity ratio K II /K I .

13.4 Disclinations There is clearly a strong connection between the concept of a dislocation and the compatibility integrals (2.10, 2.13) introduced in §2.2.1. In fact if the integral (2.10) is evaluated clockwise along a closed path S0 enclosing a dislocation with Burgers vector B, we shall obtain ∂u ds = B S0 ∂s and indeed the term ‘dislocation’ was coined by Volterra in his pioneering studies of these integral conditions. We recall that the dislocation solution can be obtained by making a cut so as to render the body simply connected, and then imposing a rigid-body relative displacement between the two faces of the cut. A related solution can be obtained by imposing a relative rigid-body rotation between the two faces, so that the integral (2.13) evaluates to a non-zero vector. In two dimensions, rotation ω is a scalar and the corresponding solution is defined by the condition u θ (r, 0) − u θ (r, 2π) = Cr .

(13.29)

By analogy with Figure 13.1, we can also interpret the solution as that in which a thin wedge of subtended angle C  1 is welded between the cut faces. From Table 9.1 we notice that the only stress function supporting a displacement discontinuity of the form (13.29) is r 2 ln(r ), and for the prescribed angle C, φ=−

μCr 2 ln(r ) . π(κ + 1)

(13.30)

However, there are two objections to using this function in isolation: There is no obvious length scale to render the argument of the logarithm dimensionless, and Table 8.1 shows that the corresponding stresses are unbounded as r → ∞.

13.4 Disclinations

231

Both these difficulties can be avoided by considering a traction-free cylinder or disk of finite radius a. The traction-free condition can be satisfied by superposing a state of uniform two-dimensional hydrostatic stress through the stress function Br 2 , and the final stress field is then defined by the non-zero stress components σrr = −



r  2μC 2μC  r  ; σθθ = − ln +1 . ln π(κ + 1) a π(κ + 1) a

This stress state is known as a disclination. For C > 0, the stress state near the origin is one of logarithically unbounded hydrostatic tension, whereas near r = a it approaches uniaxial (hoop) compression. The case C < 0 corresponds to the removal of a sector of the circle before the faces are rejoined, and involves compression near the origin and uniaxial tension near r = a. In the plane stress case, the stresses for C < 0 can be completely relaxed by allowing the disk to deform into a thin-walled cone, as is easily confirmed experimentally using a paper or cardboard disc13 . In two dimensions, dislocations and disclinations are point defects, meaning that the differential compatibility equation (2.5) is satisfied everywhere except at a singular point. However, in three dimensions they are line defects, since the cut now comprises a surface whose interior boundary is a line along which the stress field will be singular. Also, in three dimensions, the imposed relative rotation at the cut becomes a vector with three components, known as the Frank vector, by analogy with the Burgers vector.

13.4.1 Disclinations in a crystal structure Like dislocations, disclinations can be used to describe defects in an otherwise regular crystal structure, in which case the angular discontinuity C will be dictated by the defect-free structure. For example, in a hexagonal array, each molecule has six nearest neighbours, so to reproduce the cut and paste procedure described above, we would need to add or remove a 60◦ sector. Clearly, the regular structure could not be preserved distant from the disclination, and this implies that disclinations can arise in molecular arrays only as equal and opposite disclination pairs, also known as disclination dipoles. Figure 13.6 shows an otherwise regular structure where one molecule (A) has only five nearest neighbours, corresponding to a disclination with C = −π/3, whilst B has seven implying C = +π/3. However, this same structure can also be seen as a dislocation, with an extra row of molecules entering from the right and terminating at A.

13

You might also like to try the corresponding experiment for the opposite case C > 0.

232

13 Forces, Dislocations and Cracks

Fig. 13.6 A dislocation: Notice that A has only five nearest neighbours, whereas B has seven. Thus, the dislocation can be viewed as a disclination dipole.

This relationship can also be seen in the Airy stress function representation, since a dipole comprises equal and opposite disclinations separated by a small distance Δ. In the limit Δ → 0, if the strength of the disclinations increases so as to maintain CΔ constant, we recover the spatial derivative of equation (13.30) — i.e. 2μCr ln(r ) cos θ ∂φ =− ∂x π(κ + 1)

or

∂φ 2μCr ln(r ) sin θ =− , ∂y π(κ + 1)

(13.31)

where we have dropped null terms of the form r cos θ, r sin θ. But the expressions (13.31) are precisely the stress functions (13.9, 13.13) defining dislocations with Burgers vector B y and −Bx respectively. In other words, if the stress function for a unit disclination [C = 1 in equation (13.30)] is denoted by φ0 , then that for a dislocation of Burgers vector B is given by φ = i j B j

∂φ0 , ∂xi

where i j is the two-dimensional alternating tensor, 11 = 22 = 0, 12 = 1, 21 = −1.

13.5 Method of Images The method described in §13.3.2 applies strictly to the case of a crack in an infinite body, but it will provide a reasonable approximation if the length of the crack is small in comparison with the shortest distance to the boundary of a finite body or to some

13.5 Method of Images

233

other geometric feature such as another crack or an interface to a different material. However, the same methodology could be applied to other problems if we could obtain the solution for a dislocation located at an arbitrary point in the uncracked body. Closed-form solutions exist for bodies of a variety of shapes, including the traction-free half-plane, the infinite body with a traction-free circular hole and the infinite body containing a circular inclusion of a different elastic material14 . These solutions all depend on the location of image singularities at appropriate points outside the body. The nature and strength of these singularities can be chosen so as to satisfy the required boundary conditions on the boundary of the original body. Earlier treatments of the subject proceeded in an essentially ad hoc way until the required boundary conditions were satisfied, but a unified treatment for the case of a plane interface was developed by Aderogba15 . We consider the case in which the half-plane y > 0 with elastic constants μ1 , κ1 is bonded to the half-plane y < 0 with constants μ2 , κ2 . Suppose that a singularity (typically a dislocation or a concentrated force) exists somewhere in y > 0 and that the same singularity in an infinite plane would correspond to the Airy stress function φ0 (x, y) defined throughout the infinite plane. Aderogba showed that the resulting stress field in the bonded bi-material plane is defined by the stress functions φ1 (x, y) = φ0 (x, y) + L {φ0 (x, −y)} φ2 (x, y) = J {φ0 (x, y)}

y>0 y 0. The corresponding infinite plane solution is given by equation (13.13) as 2μBx r ln(r ) sin θ φ0 = , π(κ + 1) where r, θ are measured from the dislocation. Converting to Cartesian coördinates and moving the origin to the surface of the half-plane, we have  μBx (y − a)  2 ln x + (y − a)2 . φ0 (x, y) = π(κ + 1) Substituting into equation (13.33), we have

 2  4ay(y + a) μBx 2 L {φ0 (x, −y)} = − (y − a) ln x + (y + a) + 2 π(κ + 1) x + (y + a)2 and hence the complete stress function is 

 2 μBx 4ay(y + a) x + (y − a)2 φ= − 2 . (y − a) ln π(κ + 1) x 2 + (y + a)2 x + (y + a)2

The circular hole A similar procedure can be used to find the perturbation in a stress field due to the presence of the traction-free circular hole r = a. If the unperturbed field is defined by the function φ0 (r, θ), the perturbed field can be obtained as

r2 φ = φ0 (r, θ) − L 2 φ0 a



a2 ,θ r

 +

C(r 2 − 2a 2 ln(r )) , 4

(13.34)

Problems

where

235

 2   ∂ a2 r2 r2 r2 + 1 − 2 ∇2 L{·} = 2 + 1 − 2 r a a ∂r 4 a

and the constant C is defined as C = lim ∇ 2 φ0 (r, θ) . r →0

These mathematical operations are performed by the Maple and Mathematica files ‘holeperturbation’ and the reader can verify (for example) that they generate the solutions of §8.3.2 and §8.4.1, starting from the stress functions defining the corresponding unperturbed uniform stress fields.

Problems 13.1. Figure 13.7 shows a large body with a circular hole of radius a, subjected to a concentrated force F, tangential to the hole. The remainder of the hole surface is traction free and the stress field is assumed to tend to zero as r → ∞.

Fig. 13.7

(i) Find the stress components at the point B(a, θ) on the surface of the hole, due to the candidate stress function φ=

F Rψ cos ψ π

236

13 Forces, Dislocations and Cracks

in polar coördinates R, ψ centred on the point A as shown. (ii) Transform these stress components into the polar coördinate system r, θ with origin at the centre of the hole, O. Note: For points on the surface of the hole (r = a), we have R = 2a sin(θ/2), sin ψ = cos(θ/2), cos ψ = − sin(θ/2). (iii) Complete the solution by superposing stress functions (in r, θ) from Tables 8.1, 9.1 with appropriate Fourier components in the tractions and determining the multiplying constants from the conditions that (a) the surface of the hole be traction free (except at A) and (b) the displacements be everywhere single valued. 13.2. Use a method similar to that outlined in Problem 13.1 to find the stress field if the force F acts in the direction normal to the surface of the otherwise traction-free hole r = a. 13.3. A large plate is in a state of pure bending such that σx x = σx y = 0, σ yy = C x, where C is a constant. We now introduce a crack in the range −a < x < a, y = 0. (i) Find a suitable corrective solution which, when superposed on the simple bending field, will make the surfaces of the crack free of tractions. [Hint: represent the corrective solution by a distribution of climb dislocations in −a < x < a. Don’t forget the closure condition (13.19)]. (ii) Find the corresponding crack opening displacement as a function of x and show that it has an unacceptable negative value in −a < x < 0. (iii) Re-solve the problem assuming that there is frictionless contact in some range −a < x < b, where b is a constant to be determined. The dislocations will now have to be distributed only in b < x < a and b is found from a continuity condition at x = b. (Move the origin to the mid-point of the range b < x < a.) (iv) For the case with contact, find expressions for (a) the crack opening displacement, (b) the stress-intensity factor at x = a, (c) the dimension b and (d) the contact traction in −a < x < b, and hence verify that the contact inequalities are satisfied. 13.4. A state of uniform shear σx y = S, σx x = σ yy = 0 in a large block of material is perturbed by the presence of the plane crack −a < x < a, y = 0. Representing the perturbation due to the crack by a distribution of glide dislocations along the crack line, find the shear stress distribution on the line x > a, y = 0, the relative motion between the crack faces in −a < x < a, y = 0 and the mode II stress-intensity factor K II defined as  K II ≡ lim+ σ yx (x, 0) 2π(x − a) . x→a

13.5. A state of uniform general bi-axial stress σx x = Sx x , σ yy = S yy , σx y = Sx y in a large block of material is perturbed by the presence of the plane crack −a < x < a,

Problems

237

y = 0. Note that the crack will close completely if S yy < 0 and will be fully open if S yy > 0. Use the method proposed in Problem 13.4 to find the mode II stressintensity factor K II for both cases, assuming that Coulomb friction conditions hold in the closed crack with coefficient f . If the block contains a large number of widely separated similar cracks of all possible orientations and if the block fails when at any one crack 

K I2 + K II2 = K Ic ,

where the fracture toughness K Ic is a material constant, sketch the biaxial failure surface for the material — i.e. the locus of all failure points in principal biaxial stress space (σ1 , σ2 ). 13.6. By treating equation (13.30) as a Green’s function or otherwise, find the Airy stress function φ defining the stresses due to a disclination of total strength C uniformly distributed over the circle 0 ≤r < c at the centre of a traction-free circular disk of radius a [ > c ]. 13.7. Find the stress field due to a climb dislocation of unit strength located at the point (0, a) in the traction-free half-plane y > 0, using the following method:(i) Write the stress function (13.8) in Cartesian coördinates. (ii) Find the solution for a dislocation at (0, a) in the full plane −∞ < y < ∞ by making a change of origin. (iii) Superpose additional singularities centred on the image point (0, −a) to make the surface y = 0 traction free. Notice that these singularities are outside the actual body and are therefore admissible. Appropriate functions [in polar coördinates centred on (0, −a)] are C1r ln(r ) cos θ ;

C2 cos θ ; C3 sin(2θ) . r

13.8. Solve Problem 13.7 using Aderogba’s result (13.33, 13.32) for step (iii). 13.9. Find the stress field due to a concentrated force F applied in the x-direction at the point (0, a) in the half-plane y > 0, if the surface of the half-plane is traction free. Verify that the results reduce to those of §12.3 in the limit as a → 0. 13.10. Two dissimilar elastic half-planes are bonded on the interface y = 0 and have material properties as defined in §13.5. Use Aderogba’s formula (13.32) to determine the stress field due to a climb dislocation at the point (0, a) in the half-plane y > 0. In particular, find the location and magnitude of the maximum shear traction at the interface. 13.11. Use equation (13.34) to solve Problem 8.3.

Chapter 14

Thermoelasticity

Most materials tend to expand if their temperature rises and, to a first approximation, the expansion is proportional to the temperature change. If the expansion is unrestrained and the material is isotropic, all dimensions will expand equally — i.e. there will be a uniform dilatation described by ex x = e yy = ezz = αT ex y = e yz = ezx = 0 ,

(14.1) (14.2)

where α is the coefficient of linear thermal expansion and T is the temperature change. Notice that no shear strains are induced in unrestrained thermal expansion, so that a body which is heated to a uniformly higher temperature will get larger, but will retain the same shape. Thermal strains are additive to the elastic strains due to local stresses, so Hooke’s law is modified to the form νσ yy νσzz σx x − − + αT (14.3) ex x = E E E σx y (1 + ν) ex y = . (14.4) E

14.1 The Governing Equation The Airy stress function can be used for two-dimensional thermoelasticity, but the governing equation will generally include additional terms associated with the temperature field. Repeating the derivation of §4.3.1, but using (14.3) in place of (1.44), we find that the compatibility condition demands that ∂ 2 σ yy ∂ 2 σx y ∂ 2 σ yy ∂ 2 σx x ∂2 T ∂ 2 σx x ∂2 T − ν + Eα − 2(1 + ν) − ν + Eα =0 + ∂ y2 ∂ y2 ∂ y2 ∂x∂ y ∂x 2 ∂x 2 ∂x 2 © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 J. R. Barber, Elasticity, Solid Mechanics and Its Applications 172, https://doi.org/10.1007/978-3-031-15214-6_14

239

240

14 Thermoelasticity

and after substituting for the stress components from (4.6) and rearranging, we obtain ∇ 4 φ = −Eα∇ 2 T ,

(14.5)

for plane stress.

Plane strain The corresponding plane strain equations can be obtained by a similar procedure, noting that the restraint of the transverse strain ezz (14.1) will induce a stress σzz = −EαT

(14.6)

and hence additional in-plane strains ναT . Equation (14.5) is therefore modified to ∇ 4φ = −

Eα ∇2T , (1 − ν)

(14.7)

for plane strain and we can supplement the plane stress to plane strain conversions (3.10) with the relation α = α (1 + ν  ) . (14.8) Equations (14.5, 14.7) are similar in form to that obtained in the presence of body forces (7.7) and can be treated in the same way. Thus, we can seek any particular solution of (14.5) and then satisfy the boundary conditions of the problem by superposing a more general biharmonic function, since the biharmonic equation is the complementary or homogeneous equation corresponding to (14.5, 14.7).

Example As an example, we consider the case of the thin circular disk, r < a, with traction-free edges, raised to the temperature T = T0 y 2 = T0 r 2 sin2 θ , where T0 is a constant. Substituting this temperature distribution into equation (14.5), we obtain ∇ 4 φ = −2EαT0 and a simple particular solution is

(14.9)

14.2 Heat Conduction

241

φ0 = −

EαT0 r 4 . 32

The stresses corresponding to φ0 are σrr = −

EαT0 r 2 3EαT0 r 2 ; σθθ = − ; σr θ = 0 , 8 8

and the boundary r = a can be made traction free by superposing a uniform hydrostatic tension EαT0 a 2 /8, resulting in the final stress field1 σrr =

EαT0 (a 2 − r 2 ) EαT0 (a 2 − 3r 2 ) ; σθθ = ; σr θ = 0 . 8 8

14.2 Heat Conduction The temperature field might be a given quantity — for example, it might be measured using thermocouples or radiation methods — but more often it has to be calculated from thermal boundary conditions as a separate boundary-value problem. Most materials approximately satisfy the Fourier heat conduction law, according to which the heat flux per unit area q is linearly proportional to the local temperature gradient. i.e. q = −K ∇T ,

(14.10)

where K is the thermal conductivity of the material. The conductivity is usually assumed to be constant, though for real materials it depends upon temperature. However, the resulting non-linearity is only important when the range of temperatures under consideration is large. We next apply the principle of conservation of energy to a small cube of material. Equation (14.10) governs the flow of heat across each face of the cube and there may also be heat generated, Q per unit volume, within the cube due to some mechanism such as electrical resistive heating or nuclear reaction etc. If the sum of the heat flowing into the cube and that generated within it is positive, the temperature will rise at a rate which depends upon the thermal capacity of the material. Combining these arguments we find that the temperature T must satisfy the equation2 ρc 1

∂T = K ∇2T + Q , ∂t

(14.11)

It is interesting to note that the stress field in this case is axisymmetric, even though the temperature field (14.9) is not. 2 More detail about the derivation of this equation and other information about the linear theory of heat conduction can be found in the classical text H. S. Carslaw and J. C. Jaeger, Conduction of Heat in Solids, 2nd.ed., Clarendon Press, Oxford, 1959.

242

14 Thermoelasticity

where ρ, c are respectively the density and specific heat of the material, so that the product ρc is the amount of heat needed to increase the temperature of a unit volume of material by one degree. In equation (14.11), the first term on the right-hand side is the net heat flow into the element per unit volume and the second term, Q is the rate of heat generated per unit volume. The algebraic sum of these terms gives the heat available for raising the temperature of the cube. It is convenient to divide both sides of the equation by K , giving the more usual form of the heat conduction equation ∇2T =

Q 1 ∂T − , κ ∂t K

where κ=

(14.12)

K ρc

is the thermal diffusivity of the material. Thermal diffusivity has the dimensions area/time and its magnitude gives some indication of the rate at which a thermal disturbance will propagate through the body. We can substitute (14.12) into (14.5) obtaining   Q 1 ∂T − (14.13) ∇ 4 φ = −Eα κ ∂t K for plane stress.

14.3 Steady-state Problems Equation (14.13) shows that if the temperature is independent of time and there is no internal source of heat in the body (Q = 0), φ will be biharmonic. It therefore follows that in the steady-state problem without heat generation, the in-plane stress field is unaffected by the temperature distribution. In particular, if the boundaries of the body are traction free, a steady-state (and hence harmonic) temperature field will not induce any thermal stresses under plane stress conditions3 . Furthermore, if there are internal heat sources — i.e. if Q = 0 — the stress field can be determined directly from equation (14.13), using Q, without the necessity of first solving a boundary-value problem for the temperature T . This also implies that the thermal boundary conditions in two-dimensional steady-state problems have no effect on the thermoelastic stress field. All of these deductions rest on the assumption that the elastic problem is defined in terms of tractions. If some of the boundary conditions are stated in terms of 3

But we recall that the plane stress solution is an approximation. A more exact treatment of the steady-state thermoelastic problem for moderately thick plates is given in §16.4.1.

14.3 Steady-state Problems

243

displacements, constraint of the resulting thermal distortion will induce boundary tractions and the stress field will be affected, though it will still be the same as that which would have been produced by the same tractions if they had been applied under isothermal conditions. Recalling the arguments of §2.2.1, we conclude that similar considerations apply to the multiply-connected body, for which there exists an implied displacement boundary condition. In other words, multiply-connected bodies will generally develop non-zero thermal stresses even under steady-state conditions with no boundary tractions. However, the resulting stress field is essentially that associated with the presence of a dislocation in the hole and hence can be characterized by relatively few parameters4 . Appropriate conditions for multiply-connected bodies can be explicitly imposed, but in general it is simpler to obviate the need for such a condition by reverting to a displacement function representation. We shall therefore postpone discussion of thermoelastic problems for multiply-connected bodies until Chapter 23 where such a formulation is introduced.

14.3.1 Dundurs’ Theorem If the conditions discussed in the last section are satisfied and the temperature field therefore induces no thermal stress, the strains will be given by equations (14.1, 14.2). It then follows that ∂2u y ∂2u x , =− 2 ∂x ∂x∂ y because of (14.2) =−

αq y ∂ex x ∂T = −α = , ∂y ∂y K

(14.14)

from (14.1, 14.10). In view of (14.8), the corresponding result for plane strain can be written ∂2u y α(1 + ν)q y . (14.15) = 2 ∂x K In this equation, the constant of proportionality α(1+ν)/K is known as the thermal distortivity of the material and is denoted by the symbol δ. Equations (14.14, 14.15) state that the curvature of an initially straight line segment in the x-direction (∂ 2 u y /∂x 2 ) is proportional to the local heat flux across that line segment. This result was first proved by Dundurs5 and is referred to as Dundurs’ 4

See for example J. Dundurs (1974), Distortion of a body caused by free thermal expansion, Mechanics Research Communications, Vol. 1, pp. 121–124. 5 J. Dundurs loc. cit.

244

14 Thermoelasticity

Theorem. It is very useful as a guide to determining the effect of thermal distortion on a structure. Figure 14.1 shows some simple bodies with various thermal boundary conditions and the resulting steady-state thermal distortion.

Fig. 14.1 Distortion due to thermal expansion.

Notice that straight boundaries that are unheated remain straight, those that are heated become convex outwards, whilst those that are cooled become concave. The angles between the edges are unaffected by the distortion, because there is no shear strain. Since the thermal field is in the steady-state and there are no heat sources, the algebraic sum of the heat input around the boundary must be zero. Thus, although the boundary is locally rotated by the cumulative heat input from an appropriate starting point, this does not lead to incompatibility at the end of the circuit.

Problems

245

Many three-dimensional structures such as box-sections, tanks, rectangular hoppers etc., are fabricated from plate elements. If the unrestrained thermal distortions of these elements are considered separately, the incompatibilities of displacement developed at the junctions between elements permits the thermal stress problem to be described in terms of dislocations, the physical effects of which are more readily visualized. Dundurs’ Theorem can also be used to obtain some useful simplifications in twodimensional contact and crack problems involving thermal distortion6 .

Problems 14.1. A direct electric current I flows along a conductor of rectangular cross section −4a < x < 4a, −a < y < a, all the surfaces of which are traction free. The conductor is made of copper of electrical resistivity ρ, thermal conductivity K , Young’s modulus E, Poisson’s ratio ν and coefficient of thermal expansion α. Assuming the current density to be uniform and neglecting electromagnetic effects, estimate the thermal stresses in the conductor when the temperature has reached a steady state. 14.2. A fuel element in a nuclear reactor can be regarded as a solid cylinder of radius a. During operation, heat is generated at a rate Q 0 (1+ Ar 2 /a 2 ) per unit volume, where r is the distance from the axis of the cylinder and A is a constant. Assuming that the element is immersed in a fluid at pressure p and that axial expansion is prevented, find the radial and circumferential thermal stresses produced in the steady state. 14.3. The instantaneous temperature distribution in the thin plate −a < x < a, −b < y < b is defined by  2  x −1 , T (x, y) = T0 a2 where T0 is a positive constant. Find the magnitude and location of (i) the maximum tensile stress and (ii) the maximum shear stress in the plate if the edges x = ±a, y = ±b are traction free and a  b. 14.4. The half-plane y > 0 is subject to periodic heating at the surface y = 0, such that the surface temperature is T (0, t) = T0 cos(ωt) . Show that the temperature field T (y, t) = T0 e−λy cos(ωt − λy) satisfies the heat conduction equation (14.12) with no internal heat generation, provided that 6

For more details, see J. R. Barber (1980), Some implications of Dundurs’ Theorem for thermoelastic contact and crack problems, Journal of Strain Analysis, Vol. 22, pp. 229–232.

246

14 Thermoelasticity

 λ=

ω . 2k

Find the corresponding thermal stress field as a function of y, t if the surface of the half-plane is traction free. Using appropriate material properties, estimate the maximum tensile stress generated in a large rock due to diurnal temperature variation, with a maximum daytime temperature of 30o C and minimum nighttime temperature of 10o C. 14.5. The layer 0 < y < h rests on a frictionless rigid foundation at y = 0 and the surface y = h is traction free. The foundation is a thermal insulator and the free surface is subjected to the steady state heat input q y = q0 cos(mx) . Use Dundurs’ theorem to show that the layer will not separate from the foundation and find the amplitude of the sinusoidal perturbation in the free surface due to thermal distortion. 14.6. Figure 14.2 shows a hollow cylinder of internal radius a and external radius b that has been slit along a plane surface AB at θ = 0. The inner surface r = a is subjected to a heat flux qr = q0 (θ), whilst the outer radius r = b is maintained at zero temperature. All the surfaces, including the slit, are traction free. Using Dundurs’ theorem, show that in the steady state, the two surfaces at AB will rotate relative to each other so as to open an angle ψ = Qδ, where  Q=

2π

q0 (θ)adθ

0

is the total heat flux transmitted between the inner and outer surfaces andδ is the thermal distortivity defined in equation (14.15).

Fig. 14.2

Hence, using the results of §13.4 or otherwise, find the steady-state thermal stresses in a traction-free solid cylinder of radius a due to a line heat source Q at the origin r = 0.

Chapter 15

Antiplane Shear

In Chapters 3–14, we considered two-dimensional states of stress involving the inplane displacements u x , u y and stress components σx x , σx y , σ yy . Another class of two-dimensional stress states that satisfy the elasticity equations exactly is that in which the in-plane displacements u x , u y are everywhere zero, whilst the out-of-plane displacement u z is independent of z — i.e. u x = u y = 0 ; u z = f (x, y) .

(15.1)

Substituting these results into the strain-displacement relations (1.37) yields ex x = e yy = ezz = 0 and ex y = 0 ; e yz =

1∂f 2 ∂y

; ezx =

1∂f . 2 ∂x

It then follows from Hooke’s law (1.54) that σx x = σ yy = σzz = 0 and σx y = 0 ;

σ yz = μ

∂f ∂y

; σzx = μ

(15.2) ∂f . ∂x

(15.3)

In other words, the only non-zero stress components are the two shear stresses σzx , σzy and these are functions of x, y only. Such a stress state is known as antiplane shear or antiplane strain. The in-plane equilibrium equations (2.1, 2.2) are identically satisfied by the stress components (15.2, 15.3) if and only if the in-plane body forces px , p y are zero. However, substituting (15.3) into the out-of-plane equilibrium equation (2.3) yields μ∇ 2 f + pz = 0 . © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 J. R. Barber, Elasticity, Solid Mechanics and Its Applications 172, https://doi.org/10.1007/978-3-031-15214-6_15

(15.4) 247

248

15 Antiplane Shear

Thus, antiplane problems can involve body forces in the axial direction, provided these are independent of z. In the absence of body forces, equation (15.4) reduces to the Laplace equation ∇2 f = 0 .

(15.5)

15.1 Transformation of Coördinates It is often convenient to regard the two stress components on the z-plane as components of a shear stress vector τ = iσzx + j σzy = μ∇ f

(15.6)

from (15.3). The vector transformation equations (1.8) then show that the stress components in the rotated coördinate system x  , y  of Figure 1.3 are   σzx = σzx cos θ + σzy sin θ ; σzy = σzy cos θ − σzx sin θ .

Equation (15.6) can also be used to define the expressions for the stress components in polar coördinates as σzr = μ

∂f ∂r

;

σzθ =

μ∂f . r ∂θ

(15.7)

The maximum shear stress at any given point is the magnitude of the vector τ and is given by     ∂f 2 ∂f 2 + . |τ | = μ ∂x ∂y

15.2 Boundary Conditions Only one boundary condition can be imposed at each point on the boundary. For example, we might prescribe the axial displacement u z = f , or alternatively the traction ∂f , σnz = μ ∂n where n is the local outward normal to the cross section. Thus, antiplane problems reduce to the solution of the Poisson equation (15.4) [or the Laplace equation (15.5) when there is no body force] with prescribed values of f or ∂ f /∂n on the boundary. This is a classical boundary-value problem that is somewhat simpler than that involved in the solution for the Airy stress function, but essentially similar meth-

15.3 The Rectangular Bar

249

ods can be used for both. Simple examples can be defined for all of the geometries considered in Chapters 5–13. In the present chapter, we shall consider a few special cases, but a wider range of examples can be found in the problems at the end of this chapter.

15.3 The Rectangular Bar Figure 15.1 shows the cross section of a long bar of rectangular cross section 2a × b, where a  b. The short edges x = ±a are built in to rigid supports, the edge y = b is loaded by a uniform shear traction σ yz = S and the remaining edge y = 0 is traction free.

Fig. 15.1 Rectangular bar with shear loading on one edge.

Using equations (15.2, 15.3), the boundary conditions can be written in the mathematical form f = 0 ; x = ±a S ∂f = ; y=b ∂y μ ∂f =0 ; y=0. ∂y

(15.8) (15.9) (15.10)

However, as in Chapter 5, a finite polynomial solution can be found only if the boundary conditions (15.8) on the shorter edges are replaced by the weak form 

b

f (x, y)dy = 0 ;

x = ±a .

(15.11)

0

This problem is even in x and a solution can be obtained using the trial function f = C1 x 2 + C2 y 2 + C3 y + C4 . Substitution into equations (15.5, 15.9–15.11) yields the conditions

250

15 Antiplane Shear

2C1 + 2C2 = 0 S 2C2 b + C3 = μ C3 = 0 C 2 b3 C 3 b2 C1 a 2 b + + + C4 = 0 3 2 with solution C1 = −

S S S(3a 2 − b2 ) ; C2 = ; C3 = 0 ; C4 = . 2μb 2μb 6μb

(15.12)

The final solution for the stresses and displacements is therefore   S b2 Sx Sy 2 2 2 y −x +a − ; σzx = − ; σzy = , uz = 2μb 3 b b from equations (15.12, 15.1, 15.3).

15.4 The Concentrated Line Force Consider an infinite block of material loaded by a force F per unit length acting along the z-axis. This problem is axisymmetric, in contrast to that solved in §13.1, since the force this time is directed along the axis rather than perpendicular to it. It follows that the resulting displacement function f must be axisymmetric and equation (15.5) reduces to the ordinary differential equation d2 f 1 df =0, + 2 dr r dr

(15.13)

since there is no body force except at the origin. Equation (15.13) has the general solution f = C1 ln(r ) + C2 , leading to the stress field σr z =

μC1 r

; σθz = 0 ,

(15.14)

from (15.7). The constant C1 can be determined by considering the equilibrium of a cylinder of material of radius a and unit length. We obtain  2π σr z (a, θ)adθ = 0 F+ 0

and hence

15.4 The Concentrated Line Force

251

C1 = −

F . 2πμ

The corresponding stresses are then obtained from (15.14) as σr z = −

F 2πr

; σθz = 0 .

Of course, the same result could have been obtained without recourse to elasticity arguments, appealing simply to axisymmetry and equilibrium and hence the same stresses would be obtained for a fairly general non-linear or inelastic material, as long as it is isotropic. The elastic displacement is uz = f = −

F ln(r ) + C2 , 2πμ

where the constant C2 represents an arbitrary rigid-body displacement.

θ = −π

θ=

Fig. 15.2 Half-space loaded by an out-of-plane line force.

All θ-surfaces are traction free and hence the same solution can be used for a wedge of any angle loaded by a uniform force per unit length acting at and parallel to the apex, though the factor of 2π in C1 will then be replaced by the subtended wedge angle. In particular, Figure 15.2 shows the half-space −π < θ < 0 loaded by a uniformly distributed tangential force along the z-axis, for which it is easily shown that F F ln(r ) ; uz = − + C2 . (15.15) σr z = − πr πμ

252

15 Antiplane Shear

This result can be used to solve antiplane frictional contact problems analogous to those considered in Chapter 12 (see Problems 15.11, 15.12).

15.5 The Screw Dislocation The dislocation solution in antiplane shear corresponds to the situation in which the infinite body is cut on the half-plane x > 0, y = 0 and the two faces of the cut experience a relative displacement δ = Bz in the z-direction. Describing the cut space as the wedge 0 < θ < 2π, the boundary conditions for this problem can be written u z (r, 0) − u z (r, 2π) = Bz , and a suitable solution satisfying equation (15.5) is uz = −

Bz θ μBz ; σzr = 0 ; σzθ = . 2π 2πr

The corresponding material defect is known as a screw dislocation. Notice that the stress components due to a screw dislocation decay with r −1 with distance from the origin, as do those for the in-plane climb and glide dislocations discussed in §13.2. A distribution of screw dislocations can be used to solve problems involving cracks in an antiplane shear field, using the technique introduced in §13.3.

Problems 15.1. The long rectangular bar, 0 < x < a, 0 < y < b, a  b, is built in to a rigid support at x = a and loaded by a uniform shear traction σ yz = S at y = b. The remaining surfaces are free of traction. Find a solution for the displacement and stress fields, using strong boundary conditions on the edges y = 0, b. 15.2. Figure 15.3 shows the cross section of a long bar of equilateral triangular cross section of side a. The three faces of the bar are built in to rigid supports and the axis of the bar is vertical, resulting in gravitational loading pz = −ρg. Show that an exact solution for the stress and displacement fields can be obtained using a stress function of the form  √  √ √ 3a f = C(y − 3x)(y + 3x) y − . 2 Find the value of the constant C and the location and magnitude of the maximum shear stress.

Problems

253

15.3. A state of uniform shear σx z = S, σ yz = 0 in a large block of material is perturbed by the presence of a small traction-free hole whose boundary is defined by the equation r = a in cylindrical polar coördinates r, θ, z. Find the complete stress field in the block and hence determine the appropriate stress-concentration factor. 15.4. A state of uniform shear σx z = S, σ yz = 0 in a large block of material is perturbed by the presence of a rigid circular inclusion whose boundary is defined by the equation r = a in cylindrical polar coördinates r, θ, z. The inclusion is perfectly bonded to the elastic material and is prevented from moving, so that u z = 0 at r = a. Find the complete stress field in the block and hence determine the appropriate stress-concentration factor. 15.5. The body defined by a b. You will need to satisfy the boundary conditions in the strong sense on y = ±b and in the weak sense on x = ±a. Compare the solution you get at this stage with the predictions of the elementary Mechanics of Materials theory. Then superpose an appropriate series of harmonic functions (sinusoidal in y and hyperbolic in x) to obtain an exact solution. In what range of the ratio a/b is it necessary to include the series to obtain 1% accuracy in the maximum shear stress? 18.6. Find the distribution of shear stress in a hollow cylindrical bar of inner radius b and outer radius a, loaded by a shear force Fy = F acting through the origin. You will need to supplement the polynomial solution by the singular harmonic functions x/(x 2 + y 2 ) and (x 3 −3x y 2 )/(x 2 + y 2 )3 with appropriate multipliers4 . 18.7. A bar has the cross section of a sector 0 1, the limit (19.20) does not exist, but we can extract the residue by successive partial integrations. We obtain

 exp(ı(θ1 +2π))  

f 1 (ζ) f 1 (ζ)dζ f 1 (ζ)dζ

= − + ,

n (n−1) (ζ − ζ0 ) (n−1)(ζ − ζ0 ) (n−1)(ζ − ζ0 )(n−1)  exp(ıθ1 ) where we have used the same circular path of radius . The first term is zero for n = 1,   so f 1 (ζ)dζ f 1 (ζ)dζ = . (ζ − ζ0 )n (n − 1)(ζ − ζ0 )(n−1) Repeating this procedure until the final integral has only a simple pole, we obtain 

where

f 1 (ζ)dζ = (ζ − ζ0 )n



f 1[n−1] (ζ) ≡

f 1[n−1] (ζ)dζ , (n − 1)!(ζ − ζ0 ) d (n−1) f 1 (ζ) . dζ (n−1)

(19.21)

The residue at a pole of order n is therefore Res(ζ0 ) =

f 1[n−1] (ζ0 ) , (n − 1)!

(19.22)

where f 1[n−1] is defined through equations (19.21, 19.18). Suppose now that the function f (ζ) posesses several poles and we wish to determine the value of the contour integral along a path enclosing those at the points ζ1 , ζ2 , ..., ζk . One such path enclosing three poles at P1 , P2 , P3 is shown in Figure 19.2. It is clear that the contributions from the straight line segments will be self-

19.5 Line Integrals

309

P3 ●

P1 ●

Fig. 19.2 Integral path around three poles at P1 , P2 , P3 .

●

P2

cancelling, so the integral is simply the sum of those for the circular paths around each separate pole. Thus, the resulting integral will have the value  f (ζ)dζ = 2πı

k 

Res(ζ j ) .

(19.23)

j=1

This result is extremely useful as a technique for determining the values of complex line integrals. Similar considerations apply to the evaluation of contour integrals around one or more holes in a multiply-connected body. Example As a special case, we shall determine the value of the contour integral  S

dζ , ζ n (ζ − a)

where the integration path S encloses the points ζ = 0, a. The integrand has a simple pole at ζ = a and a pole of order n at ζ = 0. For the simple pole, we have Res(a) =

1 , an

from (19.20). For the pole at ζ = 0, we have f 1 (ζ) =

1 (ζ − a)

and

f 1[n−1] (ζ) =

(−1)(n−1) (n − 1)! , (ζ − a)n

so f 1[n−1] (0) = from (19.22).

(−1)(n−1) (n − 1)! (−a)n

and

Res(0) =

(−1)(n−1) 1 =− n , n (−a) a

310

19 Prelinary Mathematical Results

Using (19.23), we then have    dζ 1 1 − n =0; = 2πı n an a S ζ (ζ − a) = 2πı ;

n≥1 n=0,

(19.24) (19.25)

since if n = 0 we have only the simple pole at ζ = a.

19.5.2 The Cauchy integral theorem Changing the symbols in equation (19.19), we have  f (s)ds 1 f (ζ0 ) = , 2πı S (s − ζ0 )

(19.26)

where the path of the integral, S, encloses the point s = ζ0 . If S is chosen to coincide with the boundary of a simply-connected region Ω within which f is holomorphic, this provides an expression for f at a general point in Ω in terms of its boundary values. We shall refer to this as the interior problem, since the region Ω is interior to the contour. Notice that equation (19.19) and hence (19.26) is based on the assumption that the contour S is traversed in the anticlockwise direction. Changing the direction would lead to a sign change in the result. An equivalent statement of this requirement is that the contour is traversed in a direction such that the enclosed simply-connected region Ω always lies on the left of the path S. A related result can be obtained for the exterior problem in which f is holomorphic in the multiply-connected region Ω exterior to the closed contour S and extending to and including the point at infinity. In this case, a general holomorphic function f can be expressed as a Laurent series f = C0 +

∞  Ck k=1

ζk

,

(19.27)

where the origin lies within the hole excluded by S. For the exterior problem to be well posed, we must specify the value of f at all points on S and also at infinity ζ → ∞. We shall restrict attention to the case where f (∞) = 0 and hence C0 = 0, since the more general case is most easily handled by first solving a problem for the entire plane (with no hole) and then defining a corrective problem for the conditions at the inner boundary S. To invoke the Cauchy integral theorem, we need to construct a contour that encloses a simply-connected region including the general point ζ = ζ0 . A suitable contour is illustrated in Figure 19.3, where the path S corresponds to the boundary of the hole and S1 is a circle of sufficiently large radius to include the point ζ = ζ0 . Notice that the theorem requires that the contour be traversed in an anticlockwise direction as shown.

19.5 Line Integrals

311

S1

S3 S2 S

●

ζ = ζ0 Fig. 19.3 Contour for the region exterior to a hole S.

Equation (19.26) now gives f (ζ0 ) =

1 2πı

 S+S1 +S2 +S3

f (s)ds . (s − ζ0 )

(19.28)

In this integral, the contributions from the line segments S2 , S3 clearly cancel, leaving 1 f (ζ0 ) = − 2πı

 S

1 f (s)ds + (s − ζ0 ) 2πı

 S1

f (s)ds . (s − ζ0 )

(19.29)

The remaining terms are two separate contour integrals around S and S1 respectively in Figure 19.3. Notice that we have changed the sign in the integral around S, since S is now a closed contour and the convention demands that it be evaluated in the anticlockwise direction, which is opposite to the direction implied in (19.28) and Figure 19.3. Using (19.27) and recalling that C0 = 0, we have 

∞

S1

 f (s)ds = (s − ζ0 ) k=1

 S1

Ck ds =0, − ζ0 )

s k (s

from (19.24). Thus, the second term in (19.29) is zero and we have f (ζ0 ) = −

1 2πı

 S

f (s)ds , (s − ζ0 )

(19.30)

which determines the value of the function f at a point in the region exterior to the contour S in terms of the values on this contour.

312

19 Prelinary Mathematical Results

19.6 Solution of Harmonic Boundary-value Problems At first sight, equations (19.26, 19.30) appear to provide a direct method for finding the value of a harmonic function from given boundary data. However, if the required harmonic function is real, we meet a difficulty, since we know only the real part of the boundary data. We could write a general real harmonic function as φ = f + f¯ ,

(19.31)

in which case we know the boundary values of the function φ, but not of the separate functions f, f¯. However, we shall show in the next section that a direct solution can be developed for both the interior and exterior problems in the special case where the boundary S is circular. Furthermore, this solution can be extended to more general geometries using the technique of conformal mapping.

19.6.1 Direct method for the interior problem for a circle We consider the simply-connected region interior to the circle r = a. The boundary of this circle S is defined by s = a exp(ıθ) ; s = a exp(−ıθ) and hence ss = a 2 ; s =

a2 ; s∈S. s

(19.32)

(19.33)

Consider the problem of determining a real harmonic function φ(x, y) from real boundary data on r = a. We choose to write φ as the sum of a holomorphic function f and its conjugate as in (19.31). We then expand f inside the circle as a Taylor series ∞  Ck ζ k (19.34) f (ζ) = C0 + k=1

in which case f (s) = C 0 +

∞  C k a 2k k=1

sk

; s∈S,

using (19.33). We now apply the Cauchy operator to the boundary data, obtaining

19.6 Solution of Harmonic Boundary-value Problems

1 2πı



1 φ(s)ds = (s − ζ) 2πı =

1 2πı

 



1 f (s)ds + (s − ζ) 2πı



C0 f (s)ds + (s − ζ) 2πı

313

f (s)ds (s − ζ)

 ∞ 1  a 2k ds ds + Ck k (s − ζ) 2πı k=1 s (s − ζ)

= f (ζ) + C 0 , using (19.26) for the first two integrals and noting that every integral in the third term summation is zero in view of (19.24). Thus, we have  1 φ(s)ds f (ζ) = −C 0 + , (19.35) 2πı (s − ζ) after which the real harmonic function φ can be recovered from (19.31) except for the unknown complex constant C0 . To determine this constant, we equate (19.35) to the Taylor series (19.34) and evaluate it at ζ = 0, obtaining  1 φ(s)ds , C0 = −C 0 + 2πı s or C0 + C 0 = Alternatively, writing

1 2πı



φ(s)ds . s

s = aeıθ ; ds = aıeıθ dθ

from (19.32), we have C0 + C 0 =

1 2π



2π

φ(a, θ)dθ .

(19.36)

0

Thus, the real part of the constant C0 can be determined, but not its imaginary part and this is reasonable since if we add an arbitrary imaginary constant into f it will make no contribution to φ in (19.31).

Example Suppose that φ(a, θ) = φ2 cos(2θ), where φ2 is a real constant and we wish to find the corresponding harmonic function φ(r, θ) inside the circle of radius a. Using (19.32), we can write   φ2 s 2 φ2 (e2ıθ + e−2ıθ ) a2 = φ2 cos(2θ) = + 2 2 a2 s2 and it follows that

314

19 Prelinary Mathematical Results

1 2πı



φ2 φ(s)ds = (s − ζ) 4πa 2 ı



φ2 a 2 s 2 ds + (s − ζ) 4πı



φ2 ζ 2 ds = , s 2 (s − ζ) 2a 2

where we have used (19.26) for the first integral and the second integral is zero in view of (19.24). Using (19.35), we then have f = and

φ2 ζ¯2 f¯ = − C0 2a 2

φ2 ζ 2 − C0 ; 2a 2

φ2 (ζ 2 + ζ¯2 ) φ = f + f¯ = − C0 − C 0 . 2a 2

Evaluating the integral in (19.36), we obtain C0 + C 0 =

φ2 2π



2π

cos(2θ)dθ = 0 ,

0

and hence the required harmonic function is φ=

φ2 (ζ 2 + ζ¯2 ) φ2 r 2 cos(2θ) = , 2 2a a2

using (19.3). Of course, this result could also have been obtained by writing a general harmonic function as the Fourier series φ=

∞ 

An r n cos(nθ) +

n=0

∞ 

Bn r n sin(nθ)

n=1

and equating coefficients on the boundary r = a, but the present method, though longer, is considerably more direct.

19.6.2 Direct method for the exterior problem for a circle For the exterior problem, we again define φ as in (19.31) and apply the Cauchy integral operator to the boundary data, obtaining 1 2πı



1 φ(s)ds = (s − ζ) 2πı



1 f (s)ds + (s − ζ) 2πı



f (s)ds , (s − ζ)

(19.37)

where points on the contour are defined by (19.32). We assume that the function φ is required to satisfy the condition φ → 0, r → ∞, in which case the function f , which is holomorphic exterior to the circle, can be written as the

19.6 Solution of Harmonic Boundary-value Problems

315

Laurent expansion f (ζ) =

∞  Ck k=1

ζk

and hence

f (s) =

∞  C k sk k=1

a 2k

,

using (19.33). It follows that the second integral on the right-hand side of (19.37) is 1 2πı



 ∞ 1  Ck f (s)ds s k ds = 2k (s − ζ) 2πı k=1 a (s − ζ)

and this is zero since ζ lies outside the circle and hence the integrand defines a holomorphic function in the region interior to the circle. Using this result and (19.30) in (19.37), we obtain  1 φ(s)ds f (ζ) = − , (19.38) 2πı (s − ζ) after which φ is recovered from (19.31). The reader might reasonably ask why the method of §§19.6.1, 19.6.2 works for circular contours, but not for more general problems. The answer lies in the relation (19.33) which enables us to define the conjugate s in terms of s and the constant radius a. For a more general contour, this relation would also involve the coördinate r in s =r exp(ıθ) and this varies around the contour if the latter is non-circular.

19.6.3 The half-plane If the radius of the circle is allowed to grow without limit and we move the origin to a point on the boundary, we recover the problem of the half-plane, which can conveniently be defined as the region y > 0 whose boundary is the infinite line −∞ < x < ∞, y = 0. The contour for the integral theorem (19.35) then comprises a circle of infinite radius which lies entirely at infinity except for the line segment −∞ < x < ∞, y = 0 and we obtain f (ζ) = −C 0 +

1 2πı



∞ −∞

φ(x)d x . (x − ζ)

(19.39)

In some cases, an even more direct method can be used to solve the boundaryvalue problem. We first note that the holomorphic (and hence harmonic) function f (ζ) reduces to f (x) on y = 0. Thus, if we simply replace x by ζ in the boundary function φ(x), we shall obtain a function of ζ that reduces to φ(x) on the boundary y = 0. Unfortunately, the resulting function φ(ζ) will often not be holomorphic, either because the boundary function φ(x) is not continuously differentiable, or because

316

19 Prelinary Mathematical Results

φ(ζ) contains one or more poles in the half-plane y > 0. In these cases, the integral theorem (19.39) will always yield the correct solution. To illustrate this procedure, we shall give two examples.

Example 1 Consider the boundary-value problem φ(x, 0) = 1 ; −a < x < a = 0 ; |x| > a . Since the function is discontinuous, we must use the integral theorem (19.39), which here yields   a ζ −a dx 1 1 = −C 0 + ln . f (ζ) = −C 0 + 2πı −a (x − ζ) 2πı ζ +a The corresponding real function φ is then obtained from (19.31) and after using (19.1) and simplifying, we obtain      y y 1 φ(x, y) = arctan − arctan , π x −a x +a where the real constant C0 +C 0 can be determined to be zero by equating φ(x, y) to its value at any convenient point on the boundary. Example 2 Consider now the case where φ(x, 0) =

1 . (1 + x 2 )

This function is continuous and differentiable, so a first guess would be to propose the function 1 φ =  ( f 1 (ζ)) , where f 1 (ζ) = (1 + ζ 2 ) which certainly tends to the correct value on y = 0. However, f 1 has poles at ζ = ±ı, i.e. at the points (0, 1), (0, −1), the former of which lies in the half-plane y > 0. Thus f 1 (ζ) is not holomorphic throughout the half-plane and its real part is therefore not harmonic at (0, 1). As in the previous example, we can resolve the difficulty by using the integral theorem. However, the fact that the boundary function is continuous permits us to

19.7 Conformal Mapping

317

evaluate the integral using the residue theorem. We therefore write f (ζ) = −C 0 +

1 2πı

 S

ds 1 = −C 0 + (1 + s 2 )(s − ζ) 2πı

 S

ds , (s + ı)(s − ı)(s − ζ)

where the contour S is the circle of infinite radius which completely encloses the half-plane y > 0. The integrand has simple poles at s = ı, −ı, ζ, but only the first and third of these lie within the contour. We therefore obtain  1 1 1 ı ds = + = 2πı S (s + ı)(s − ı)(s − ζ) (ζ + ı)(ζ − ı) 2ı(ı − ζ) 2(ζ + ı) and the real harmonic function φ is then recovered as ¯ = f + f¯ − C0 − C 0 = φ(ζ, ζ)

ı ı − − C0 − C 0 . ¯ 2(ζ + ı) 2(ζ − ı)

The constants can be determined by evaluating this expression at the origin ζ = 0 and equating it to φ(0, 0), giving 1 1 + − C0 − C 0 = 1 2 2

or

C0 + C 0 = 0 .

¯ in terms of (x, y), obtaining We then use (19.1) to express φ(ζ, ζ) φ(x, y) =

x2

(1 + y) . + (1 + y)2

This function can be verified to be harmonic and it clearly tends to the required function φ(x, 0) on y = 0.

19.7 Conformal Mapping In §19.6, we showed that the value of a real harmonic function in the region interior or exterior to a circle or in the half-plane can be written down as an explicit integral of the boundary data, using the Cauchy integral theorem. The usefulness of this result is greatly enhanced by the technique of conformal mapping, which permits us to ‘map’ a holomorphic function in the circular domain or the half-plane to one in a more general domain, usually but not necessarily with the same connectivity. Suppose that ω(ζ) is a holomorphic function of ζ in a domain Ω, whose boundary S is the unit circle |ζ| = 1. Each point (x, y) in Ω corresponds to a unique value ω = ξ + ıη,

318

19 Prelinary Mathematical Results

whose real and imaginary parts (ξ, η) can be used as Cartesian coördinates to define a point in a new domain Ω ∗ with boundary S ∗ . Thus the function ω(ζ) can be used to map the points in Ω into the more general domain Ω ∗ and points on the boundary S into S ∗ . Since ζ ∈ Ω and ω ∈ Ω ∗ , we shall refer to Ω, Ω ∗ as the ζ-plane and the ω-plane respectively. Next suppose that f (ω) is a holomorphic function of ω in Ω ∗ , implying that both the real and imaginary parts of f are harmonic functions of (ξ, η), from (19.10). Since ω is a holomorphic function of ζ, it follows that f ζ ≡ f (ω(ζ)) is a holomorphic function of ζ in Ω and hence that its real and imaginary parts are also harmonic functons of (x, y). Thus, the mapping function ω(ζ) maps holomorphic functions in Ω ∗ into holomorphic functions in Ω. It follows that a harmonic boundary-value problem for the domain S ∗ can be mapped into an equivalent boundary-value problem for the unit circle, which in turn can be solved using the direct method of §19.6.1 or §19.6.2.

Example: The elliptical hole A simple illustration is provided by the mapping function   m , ω ≡ ξ + ıη = c ζ + ζ

(19.40)

where c, m are real constants and m < 1. Substituting ζ =r exp(ıθ) into (19.40) and separating real and imaginary parts, we find m

m

cos θ ; η = c r − sin θ . ξ=c r+ r r Eliminating θ between these equations, we then obtain

where

ξ2 η2 + =1, α2 β2

(19.41)

m

m

; β=c r− . α=c r+ r r

(19.42)

Equation (19.41) with (19.42) shows that the set of concentric circles (lines of constant r ) in the ζ-plane maps into a set of confocal ellipses in the ω-plane. In particular, the unit circle r = 1 maps into the ellipse with semi-axes a = c(1 + m) ; b = c(1 − m) .

Problems

319

If f (ω) is a holomorphic function in the region interior to the ellipse, the mapped function f ζ will generally be unbounded both at ζ = 0 and at ζ → ∞. This mapping cannot therefore be used for the interior problem for the ellipse. It can however be used for the exterior problem, since in this case, f (ω) must posess a Laurent series, every term of which maps into a bounded term in f ζ . Equation (19.40) can be solved to give ζ as a function of ω. We obtain the two solutions √ ω ± ω 2 − 4mc2 . (19.43) ζ= 2c The positive square root maps points outside the unit circle into points outside the ellipse, whereas the negative root maps points inside the unit circle into points outside the ellipse. Either mapping can be used when combined with the appropriate solution of the boundary-value problem. Thus we can use §19.6.1 and the negative root in (19.43), or §19.6.2 and the positive root. The present author favours the exterior-to-exterior mapping implied by the positive root, since this preserves the topology of the original problem and hence tends to a more physically intuitive procedure. We then have ζ=

ω+

√

ω 2 − 4mc2 , 2c

(19.44)

and it is clear that circles of radius r > 1 in the ζ-plane map to increasingly circular contours in the ω-plane as r → ∞. To solve a given problem, we map the boundary conditions to the unit circle in the ζ-plane using (19.40), solve for the required holomorphic function of ζ using §19.6.2 and then map this function back into the ω-plane using (19.44).

Problems 19.1. By writing ζ n = en ln(ζ) , show that

∂m  n  ζ = ζ n lnm (ζ) . ∂n m

Using equations (19.3), find the real and imaginary parts of the function ζ n ln2 (ζ) in polar coördinates r, θ and verify that they are both harmonic. 19.2. Verify that the real stress function φ =r 2 θ satisfies the biharmonic equation using (8.11). Then express it in the complex form (19.15). 19.3. Express the real biharmonic function φ =r θ sin θ (from equation (8.31) or Table 8.1) in the complex form (19.15).

320

19 Prelinary Mathematical Results

19.4. Find the imaginary part of the holomorphic function whose real part is the harmonic function ln(r ). 19.5. Use the residue theorem to evaluate the contour integral  S

ζ n dζ , (ζ 2 − a 2 )

where n is a positive integer and the contour S is the circle |ζ| = 2a. 19.6. Find the real and imaginary parts of the function sin(ζ) and hence find the points in the x, y-plane at which sin(ζ) = 0. Use the residue theorem to evaluate the integral  S

dζ sin(ζ)

for a contour S that encloses only the pole at ζ = 0. By choosing S as a rectangle two of whose sides are the lines x = ±π/2, −∞ < y < ∞, use your result to evaluate the real integral ∞ dy . cosh y −∞ 19.7. Show that on the unit circle,   1 1 ζ+ = cos θ ; 2 ζ

  1 n 1 ζ + n = cos(nθ) . 2 ζ

Use this result and the notation ω = eıφ to express the real integral (12.22) in terms of an integral around the contour shown in Figure 19.4. Use the residue theorem to evaluate this integral and hence prove equation (12.22). Explain in particular how you determine the contribution from the small semi-circular paths around the poles ζ = ω, ζ = 1/ω.

S

1 0

●

ω

●

●

Fig. 19.4

19.8. Use the change of variable ξ = a cos θ and ζ = eıθ to show that

1/ω

Problems

321



1 −1

ξdξ ı  = 2 2 (x − ξ) 1 − ξ



2π 0

(1 + ζ 2 )dζ , ζ(ζ 2 − 2xζ + 1)

where the contour integral is evaluated around the unit circle. Find the location of the poles of the integrand assuming x is real and greater than unity. Hence, use the residue theorem to evaluate the integral and confirm the expression (13.20). 19.9. Find the real harmonic function φ(r, θ) defined inside the circle 0 ≤r < a whose boundary values are φ(a, θ) = 1 ; −α < θ < α = 0 ; α < |θ| < π . 19.10. Find the real harmonic function φ(r, θ) defined in the external region r > a satisfying the boundary conditions φ(a, θ) = φ3 sin(3θ) ; φ(r, θ) → 0 ; r → ∞ , where φ3 is a real constant. 19.11. Find the real harmonic function φ(x, y) in the half-plane y ≥ 0 satisfying the boundary conditions φ(x, 0) =

1 ; φ(x, y) → 0 ; (1 + x 2 )2



x 2 + y2 → ∞ .

19.12. Use the mapping function (19.40) to find the holomorphic function f (ω) such that the real harmonic function φ = f + f¯ satisfies the boundary condition φ = Aξ on the ellipse ξ2 η2 + =1 a2 b2 and f → 0 as |ω| → ∞, where A is a real constant. Do not attempt to express φ as an explicit real function of ξ, η. 19.13. Show that the function

ω(ζ) = ζ 1/2

maps the quarter plane ξ ≥ 0, η ≥ 0 into the half-plane y > 0. Use this result and appropriate results from the example in §19.6.3 to find the real harmonic function φ(ξ, η) in the quarter plane with boundary values φ(ξ, 0) =

1 ; ξ>0, 1 + ξ4

φ(0, η) =

1 ; η>0. 1 + η4

Chapter 20

Application to Elasticity Problems

20.1 Representation of Vectors From a mathematical perspective, both two-dimensional vectors and complex numbers can be characterized as ordered pairs of real numbers. It is therefore a natural step to represent the two components of a vector function V by the real and imaginary parts of a complex function. In other words, V ≡ i Vx + j Vy is represented by the complex function V = Vx + ı Vy .

(20.1)

In the same way, the vector operator ∇≡i

∂ ∂ ∂ ∂ ∂ + j = +ı =2 , ∂x ∂y ∂x ∂y ∂ ζ¯

(20.2)

from equation (19.6). If two vectors f , g are represented in the form f = fx + ı f y ;

g = gx + ıg y

the product   f¯g + f g = ( f x − ı f y )(gx + ıg y ) + ( f x + ı f y )(gx − ıg y ) = 2 f x gx + f y g y . It follows that the scalar (dot) product

Supplementary Information The online version contains supplementary material available at https://doi.org/10.1007/978-3-031-15214-6_20.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 J. R. Barber, Elasticity, Solid Mechanics and Its Applications 172, https://doi.org/10.1007/978-3-031-15214-6_20

323

324

20 Application to Elasticity Problems

f ·g=

 1 ¯ fg+ fg . 2

(20.3)

This result can be used to evaluate the divergence of a vector field V as div V ≡ ∇·V =

∂V ∂V + , ∂ζ ∂ ζ¯

(20.4)

using (20.2, 20.3). We also record the complex-variable expression for the vector (cross) product of two in-plane vectors f , g, which is  ı  (20.5) f g − f¯g k , 2 where k is the unit vector in the out-of-plane direction z. It then follows that   ∂V ∂V curl V ≡ ∇×V = ı − k, (20.6) ∂ζ ∂ ζ¯ f × g = ( f x g y − f y gx )k =

from (20.2, 20.5). Also, the moment of a force F = Fx + ı Fy about the origin is  ı  ζ F − ζ¯ F k , (20.7) 2 where ζ = x + ı y represents a position vector defining any point on the line of action of F. M=r×F=

20.1.1 Transformation of coördinates If a vector V = Vx + ı Vy is transformed into a new coördinate system (x  , y  ) rotated anticlockwise through an angle α with respect to x, y, the transformed components can be combined into the complex quantity Vα = Vx + ı Vy , where

Vx = Vx cos α + Vy sin α ;

(20.8)

Vy = Vy cos α − Vx sin α ,

from (1.8). Substituting these expressions into (20.8), we obtain Vα = (cos α − ı sin α)Vx + (sin α + ı cos α)Vy = e−ıα V .

(20.9)

In describing the tractions on a surface defined by a contour S, it is sometimes helpful to define the unit vector Δs ˆt = |Δs|

20.2 The Antiplane Problem

325

in the direction of the local tangent, where s is a position vector defining a point on S. If the contour is traversed in the conventional anticlockwise direction, the outward normal from the enclosed area nˆ is then defined by a unit vector rotated through π/2 clockwise from ˆt and hence nˆ = ˆt ×k

nˆ = tˆe−ıπ/2 = −ı tˆ ,

or

ˆ ˆt where n, ˆ tˆ are complex numbers of unit magnitude representing the unit vectors n, respectively. For the special case where s represents the circle s = a exp(ıθ), we have tˆ =

aı exp(ıθ)Δθ = ı exp(ıθ) aΔθ

nˆ = exp(ıθ) .

and

20.2 The Antiplane Problem To introduce the application of complex-variable methods to elasticity, we first consider the simpler antiplane problem of Chapter 15 and restrict attention to the case where there is no body force ( pz = 0). The only non-zero displacement component is u z , which is constant in direction and hence a scalar function in the x y-plane. From equations (15.1, 15.5), we also deduce that u z is a real harmonic function, and we can satisfy this condition by writing 2μu z = h + h ,

(20.10)

where h is a holomorphic function of ζ and h is its conjugate. The non-zero stress components σzx , σzy can be combined into a vector which we denote as  Ψ ≡ σzx + ıσzy = μ

∂u z ∂u z +ı ∂x ∂y

 = 2μ

∂u z , ∂ ζ¯

(20.11)

from (15.3, 19.6). Substituting (20.10) into (20.11) and noting that h is independent ¯ we then obtain of ζ, (20.12) Ψ = h . We can write the complex function h = hx + ı h y , where h x , h y are real harmonic functions representing the real and imaginary parts of h. Equation (20.12) can then be expanded as Ψ =

1 2



∂ ∂ +ı ∂x ∂y

 (h x − ı h y ) =

1 2



∂h y ∂h y ∂h x ∂h x + +ı −ı ∂x ∂y ∂y ∂x



326

20 Application to Elasticity Problems

and hence σzx =

1 2



∂h y ∂h x + ∂x ∂y

 ; σzy =

1 2



∂h y ∂h x − ∂y ∂x

 .

However, the Cauchy-Riemann relations (19.7) show that ∂h y ∂h x = ; ∂x ∂y

∂h y ∂h x =− ∂y ∂x

so these expressions can be simplified as σzx =

∂h y ∂h x = ; ∂x ∂y

σzy =

∂h y ∂h x =− . ∂y ∂x

Comparison with equations (17.4) then shows that h y is identical with the real Prandtl stress function ϕ — i.e. h−h =ϕ. h y ≡ (h) = 2ı It follows as in equation (17.5) that the boundary traction σzn =

∂h y , ∂t

(20.13)

where t is a real coördinate measuring distance around the boundary in the anticlockwise direction. Also, from (20.10), we have hx =

h+h = μu z = μ f (x, y) , 2

where f (x, y) is the function introduced in equation (15.1). Thus, h = h x + ı h y = μ f + ıϕ , showing that the real functions μ f and ϕ from Chapters 15, 17 can be combined as the real and imaginary parts of the same holomorphic function, for antiplane problems without body force.

20.2.1 Solution of antiplane boundary-value problems If the displacement u z is prescribed throughout the boundary, equation (20.10) reduces the problem to the search for a holomorphic function h such that h + h takes specified values on the boundary. If instead the traction σzn is specified everywhere on the boundary, we must first integrate equation (20.13) to obtain the boundary values of h y and the problem then reduces to the search for a holomorphic function

20.2 The Antiplane Problem

327

h such that h −h takes specified values on the boundary. This is essentially the same boundary-value problem, since we can define a new holomorphic function g through the equations ıh (20.14) h y = g + g with g = − . 2 Example As an example, we consider Problem 15.3 in which a uniform antiplane stress field σzx = S is perturbed by the presence of a traction-free hole of radius a. As in §8.3.2 and §8.4.1, we start with the unperturbed solution in which σzx = S everywhere. From (20.11, 20.12), Ψ = h = S , and hence a particular solution (omitting a constant of integration which corresponds merely to a rigid-body displacement) is h = S ζ¯ ;

h = Sζ ,

where we note that S is a real constant, so S = S. In particular, h y = (h) = Sy = Sr sin θ . To make the boundary of the hole traction free, we require h y to be constant around the hole and this constant can be taken to be zero without loss of generality. Thus, the corrective solution must satisfy h y (a, θ) = −Sa sin θ . It is possible to ‘guess’ the form of the solution, following arguments analogous to those used for the in-plane problem in §8.3.2, but here we shall illustrate the direct method by using equation (19.38). On the boundary r = a, we have s = a exp(ıθ) ;

sin θ =

1 s a (eıθ − e−ıθ ) = − . 2ı 2ı a s

(20.15)

Since this is an exterior problem, we use (19.38) with (20.14, 20.15) to obtain −

ıh 1 Sa = 2 2πı 2ı

or h=

S 2πı

 a ds s − , a s (s − ζ)



 a2 ds s− . s (s − ζ)

We can use the residue theorem (19.23) to compute the contour integral. Since ζ is outside the contour, only the pole a 2 /s at s = 0 makes a contribution and we obtain

328

20 Application to Elasticity Problems

Res(0) =

Sa 2 2πıζ

giving

h=

Sa 2 . ζ

The complete stress function (unperturbed + corrective solution) is then h = Sζ +

Sa 2 ζ

and the stress field is obtained from (20.12) as Ψ = h = S −

Sa 2 . ζ¯2

As a check on this result, we note that on ζ = s, h = Ss +

Sa 2 = S(s + s) , s

using a 2 = ss from (19.33). Thus, on the boundary, h is the sum of a complex quantity and its conjugate which is real, confirming that the imaginary part h y = 0 as required.

20.3 In-plane Deformations We next turn our attention to the problem of in-plane deformation, which was treated using the Airy stress function in Chapters 4–13. Our strategy will be to express the in-plane displacement vector u = u x + ıu y in terms of complex displacement functions, much as the stresses in Chapter 4 were expressed in terms of the real Airy function φ. Also, we assume plane strain conditions, so that the out of plane displacement u z = 0. By choosing to represent the displacements rather than the stresses, we automatically satisfy the compatibility conditions, as explained in §2.2, but the stresses must then satisfy the equilibrium equations and this will place restrictions on our choice of displacement functions. In vector notation, the equilibrium equations in terms of displacements require that (1 − 2ν) p =0, ∇div u + (1 − 2ν)∇ 2 u + μ from (2.16). Using (20.3, 20.2, 20.4, 19.9), we can express the two in-plane components of this condition in the complex-variable form

20.3 In-plane Deformations

∂ 2 ∂ ζ¯



∂u ∂u + ∂ζ ∂ ζ¯

329

 + 4(1 − 2ν)

where the in-plane body force

∂2u

∂ζ∂ ζ¯

+

(1 − 2ν) p =0, μ

(20.16)

p = px + ı p y .

(20.17)

¯ we obtain Integrating (20.16) with respect to ζ, ∂u (1 − 2ν) ∂u + = f − (3 − 4ν) ¯ ∂ζ 2μ ∂ζ



pd ζ¯ ,

(20.18)

where f is an arbitrary holomorphic function of ζ. This complex equation really comprises two separate equations, one for the real part and one for the imaginary part, both of which must be satisfied by u. It follows that the conjugate equation  (1 − 2ν) ∂u ∂u ¯ = f − pdζ (20.19) + (3 − 4ν) ∂ζ 2μ ∂ ζ¯ must also be satisfied. We can then eliminate ∂u/∂ ζ¯ between (20.18, 20.19) to obtain ∂u (1 − 2ν) 8(1 − 2ν)(1 − ν) = (3 − 4ν) f − f¯ + ∂ζ 2μ



 pdζ − (3 − 4ν)

¯ pd ζ ,

with solution

 8(1 − 2ν)(1 − ν)u = (3 − 4ν) f dζ − ζ f¯ + g

   (1 − 2ν) ¯ + pdζ − (3 − 4ν) pd ζ dζ , (20.20) 2μ

¯ where g is an arbitrary function of ζ.  The functions f , g are arbitrary and, in particular, f dζ is simply a new arbitrary function of ζ. We can therefore write a more compact form of (20.20) by defining different arbitrary functions through the relations  4(1 − ν)(1 − 2ν)θ 4(1 − ν)(1 − 2ν)χ ; g=− , f dζ = μ μ where χ, θ are holomorphic functions of ζ and ζ¯ only respectively. We then have 4(1 − ν)(1 − 2ν)χ , f¯ = μ

(20.21)

where χ represents the derivative of χ (which is a function of ζ¯ only) with respect ¯ to ζ.

330

20 Application to Elasticity Problems

With this notation, equation (20.20) takes the form 2μu = (3 − 4ν)χ − ζ

χ

1 −θ+ 8(1 − ν)

 

 pdζ − (3 − 4ν)

¯ pd ζ dζ .

From this point on, we shall restrict attention to the simpler case where the body forces are zero ( p = p = 0), in which case1 2μu = (3 − 4ν)χ − ζ χ − θ .

(20.22)

It is a simple matter to reintroduce body force terms in the subsequent derivations if required.

20.3.1 Expressions for stresses The tractions on the x and y-planes are τx = σx x + ıσx y ; τ y = σ yx + ıσ yy

(20.23)

and these can be combined to form the functions Φ ≡ τx + ıτ y = σx x + 2ıσx y − σ yy

(20.24)

Θ ≡ τx − ıτ y = σx x + σ yy .

(20.25)

We note that Θ is a real function and is invariant with respect to coördinate transformation, whilst Φ transforms according to the rule Φα ≡ σx x + 2ıσx y − σ yy = e−2ıα Φ ,

(20.26)

where Φα and the stress components σ  are defined in a coördinate system x  , y  rotated anticlockwise through an angle α with respect to x, y. This result is easily established using the stress-transformation equations (1.9–1.11). Substituting (20.22) into (20.4) and recalling that the state is one of plane strain so u z = 0, we find   2μ div u = 2(1 − 2ν) χ + χ and hence, using the stress-strain relations (1.54) 1

This equation can also be written in the form 2μu = κχ − ζ χ − θ ,

where κ is Kolosov’s constant defined in equation (3.11). This has the advantage of unifying the plane strain and plane stress formulations.

20.3 In-plane Deformations

331

Θ = 2(λ + μ) div u = 2(χ + χ )   ∂u y ∂u y ∂u x ∂u x Φ = 2μ − +ı + ∂x ∂y ∂y ∂x ∂u = 4μ = −2(ζ χ + θ ) . ∂ ζ¯

(20.27)

(20.28)

These expressions can also be written in terms of the single complex function ¯ = χ + ζχ + θ . ψ(ζ, ζ) We then have

∂ψ = χ + χ ; ∂ζ

and hence Θ=2

(20.29)

∂ψ = ζ χ + θ ∂ ζ¯

∂ψ ∂ψ . ; Φ = −2 ∂ζ ∂ ζ¯

(20.30)

Notice however that ψ depends on both ζ and ζ¯ and is therefore not holomorphic.

20.3.2 Rigid-body displacement Equations (20.30) show that all the stress components will be zero everywhere if and ¯ is a constant. Expanding the holomorphic functions χ, θ as Taylor only if ψ(ζ, ζ) series χ = A0 + A1 ζ + A2 ζ 2 + ... ; θ = B0 + B1 ζ + B2 ζ 2 + ... and substituting into (20.29), we obtain ¯ + B¯ 2 ζ¯2 + ... . ψ = A0 + B¯ 0 + (A1 + A¯ 1 )ζ + B1 ζ¯ + A2 ζ 2 + 2 A¯ 2 ζζ This will be constant if and only if (A1 + A¯ 1 ) = 0 and all the other coefficients are zero except A0 , B¯ 0 . The condition (A1 + A¯ 1 ) = 0 implies that A1 is pure imaginary, which can be enforced by writing A1 = ıC1 , where C1 is a real constant. Substituting these values into (20.22), we then obtain 2μu = (3 − 4ν)A0 − B¯ 0 + 4(1 − ν)ıC1 ζ .

(20.31)

The first two terms correspond to an arbitrary rigid-body translation and the third to a rigid-body rotation. In fact, either of the complex constants A0 , B¯ 0 has sufficient degrees of freedom to define an arbitrary rigid-body translation, so one of them can be

332

20 Application to Elasticity Problems

set to zero without loss of generality. In the following derivations, we shall therefore generally set B0 = B¯ 0 = 0, implying that θ(0) = 0.

20.4 Relation between the Airy Stress Function and the Complex Potentials Before considering methods for solving boundary-value problems in complexvariable notation, it is of interest to establish some relationships with the Airy stress function formulation of Chapters 4–13. In particular, we shall show that it is always possible to find complex potentials χ, θ corresponding to a given Airy function φ. This has the incidental advantage that the resulting complex potentials can then be used to determine the displacement components, so this procedure provides a method for generating the displacements due to a given Airy stress function, which, as we saw in Chapter 9, is not always straightforward in the real stress function formulation. We start by using the procedure of §19.4.1 to express the real biharmonic function φ in terms of two holomorphic functions g1 , g2 , as in equation (19.15). The stress components are given in terms of φ as σx x =

∂2φ ∂2φ ∂2φ ; σ ; σ = − = , x y yy ∂ y2 ∂x∂ y ∂x 2

from equations (4.6). It follows that Θ = ∇2φ = 4

∂2φ ∂ζ∂ ζ¯

(20.32)

and   ∂2φ ∂ 2 ∂2φ ∂ ∂2φ ∂2φ − + ı − 2ı = − = −4 , Φ= ∂ y2 ∂x∂ y ∂x 2 ∂x ∂y ∂ ζ¯2

(20.33)

using (19.9, 19.6). Substituting for φ from (19.15) into (20.32), we then have  Θ = ∇ 2 φ = 4 g2 + g2 and comparison with (20.27) shows that χ = 2g2 . Similarly, from (19.15, 20.33) we have   Φ = −4 g1 + ζ g2 = −2 ζ χ + 2g1 ,

(20.34)

20.4 Relation between the Airy Stress Function and the Complex Potentials

333

using (20.34). Comparison with (20.28) then shows that θ = 2g1 .

(20.35)

Example As an example, we consider the problem of §5.2.1 in which the rectangular beam x > 0, −b < y < b is loaded by a shear force F at the end x = 0. The real stress function for this case is given by (5.21) as φ=

F(x y 3 − 3b2 x y) . 4b3

Using (19.2), we have xy = −

ı  2 ¯2  ζ −ζ 4

and

x y3 =

2 ı  2 ¯2   ζ − ζ ζ − ζ¯ , 16

from which φ=

 ıF  4 ζ − 2ζ 3 ζ¯ + 2ζ ζ¯3 − ζ¯4 + 12b2 ζ 2 − 12b2 ζ¯2 3 64b

Comparison with (19.15) shows that g1 =

 ıF  4 ζ + 12b2 ζ 2 ; 64b3

We then have χ=−

g2 = −

ı Fζ 3 . 32b3

 ı Fζ 3 ıF  ; θ = 3 ζ 3 + 6b2 ζ , 3 16b 8b

(20.36)

from (20.34, 20.35), and hence 2μu = 2μ(u x + ıu y ) = (3 − 4ν)χ − ζ χ − θ  ıF  = −(3 − 4ν)ζ 3 − 3ζ ζ¯2 + 2ζ¯3 + 12b2 ζ¯ 3 16b    F  , = 3 3(1−ν)x 2 y −(2−ν)y 3 +3b2 y +ı 3b2 x −3νx y 2 −(1−ν)x 3 4b agreeing with the plane-strain equivalents of (9.8, 9.9) apart from an arbitrary rigidbody displacement.

334

20 Application to Elasticity Problems

20.5 Boundary Tractions We denote the boundary traction (i.e. the force per unit length along the boundary S) as T (s), where s is a real coördinate defining position around S. The traction is a vector which can be expressed in the usual complex form as T (s) ≡ i Tx + j Ty = Tx + ı Ty ≡ T . Following the conventions of Chapter 19, the coördinate s increases as we traverse the boundary in the anticlockwise direction as shown in Figure 20.1(a), which also shows the traction components Tx , Ty . Figure 20.1(b) shows the forces acting on

σ yy dx σ yx dx

Tx s

Ty

σ xxdy σ xy dy

(a)

Tx ds Ty ds (b)

Fig. 20.1 (a) The boundary tractions Tx , Ty and (b) equilibrium of a small element at the boundary.

a small triangular element chosen such that x, y, s all increase as we move up the inclined edge2 . We recall the convention that the material lies always on the left of the boundary as s increases. Equilibrium of this element then requires that Tx ds = σx x dy − σ yx d x ;

Ty ds = σx y dy − σ yy d x

and hence T ds = (σx x + ıσx y )dy − (σ yx + ıσ yy )d x = τx dy − τ y d x , using the notation of equation (20.23). We can solve (20.24, 20.25) for τx , τ y , obtaining τx = 2

1 ı (Θ + Φ) ; τ y = (Θ − Φ) 2 2

If any other orientation were chosen, one or more of the increments d x, dy, ds would be negative resulting in a change in direction of the corresponding forces, but the same final result would be obtained.

20.5 Boundary Tractions

335

and hence T ds =

  1 ı  Θ(dy − ıdx) + Φ(dy + ıdx) = Φd ζ¯ − Θdζ . 2 2

(20.37)

Using (20.30), we then have   ∂ψ ¯ ∂ψ T ds = ı − dζ − dζ = −ıdψ . ∂ζ ∂ ζ¯

(20.38)

In other words, T = −ı

d  dψ = −ı χ + ζ χ + θ ds ds

dψ = ıT . ds

or

(20.39)

Thus, if the tractions are known functions of s, equation (20.39) can be integrated to yield the values of ψ at all points on the boundary. We also note that ψ must be constant in any part of the boundary that is traction free.

Example To illustrate these results, we revisit the example in §20.4 in which the rectangular bar −b < y < b, x > 0 is loaded by a transverse force F on the end x = 0, the edges y = ±b being traction free. The complex potentials are given in (20.36). Substituting them into (20.29), we find ψ=−

  ıF  ¯ 3 + 3ζ¯ (ζ − ζ) ¯ 2 + 4b2 (ζ − ζ) . 16b3

(20.40)

In complex coördinates, the boundaries y = ±b correspond to the lines ζ − ζ¯ = ±2ıb , from (19.2) and it follows that the second term in the braces in (20.40) is zero on each boundary, whilst the first term takes the value ∓8ıb3 ,

corresponding to

ψ=∓

F . 2

This confirms that ψ is constant along each of the two traction-free boundaries. We also note that as we go down the left end of the bar (thereby preserving the anticlockwise direction for s), ψ increases by F, so  F=

ζ=−ıb ζ=ıb

dψ ds = ds



ζ=−ıb ζ=ıb

ı T ds ,

336

20 Application to Elasticity Problems

using (20.39). It follows that the resultant force on the end is 

ζ=−ıb ζ=ıb

T ds = −ı F ,

which corresponds to a force F in the negative y-direction, which of course agrees with Figure 5.2 and equation (5.14).

20.5.1 Equilibrium considerations The in-plane problem is well posed if and only if the applied loads are selfequilibrated and since we are assuming that there are no body forces, this implies that the resultant force and moment of the boundary tractions must be zero. The force resultant can be found by integrating the complex traction T around the boundary S, using (20.39). We obtain



Fx + ı Fy =

T ds = −ı S

S

dψ ds = 0 , ds

(20.41)

which implies that ψ must be single-valued. This condition applies regardless of the shape of the boundary, assuming the body is simply connected. The contribution to the resultant moment from the traction T ds on an element of boundary ds can be written dM =

 ı  1 sT ds − sT ds = − sdψ + sdψ = − (sdψ) , 2 2

using (20.7) and (20.38). The resultant moment can then be written



d M = −

M= S

s S

∂ψ ds ∂s

(20.42)

and integrating by parts, we have

s S

∂ψ ds = sψ| S − ∂s



ψds = − S

ψds , S

since sψ is single-valued in view of (20.41). Thus, the condition that the resultant moment be zero reduces to  

ψds = 0 . (20.43) M = S

20.6 Boundary-value Problems

337

20.6 Boundary-value Problems Two quantities must be specified at every point on the boundary S. In the displacement boundary-value problem, both components of displacement are prescribed and hence we know the value of the complex function 2μu(s) = κχ(s) − s χ (s) − θ(s) ; s ∈ S , where we recall that κ = (3 − 4ν). In the traction boundary-value problem, both components of traction T (s) are prescribed and we can integrate equation (20.39) to obtain  ψ(s) = ı T (s)ds + C , (20.44) where C is an arbitrary constant of integration. The complex potentials must then be chosen to satisfy the boundary condition χ(s) + s χ (s) + θ(s) = ψ(s) ; s ∈ S . Notice that the variable of integration s in (20.44) is a real curvilinear cöordinate measuring the distance traversed around the boundary S, but since we shall later want to apply the Cauchy integral theorem, it is necessary to express the resulting function ψ as a function of the complex cöordinate s defining a general point on S. The displacement and traction problems both have the same form and can be combined as (20.45) γχ(s) + s χ (s) + θ(s) = f (s) ; s ∈ S , where γ = −κ, f (s) = −2μu(s) for the displacement problem and γ = 1, f (s) = ψ(s) for the traction problem. One significant difference between the displacement and traction problems is that the former is completely defined for all continuous values of the boundary data, whereas the solution of traction problem is indeterminate to within an arbitrary rigidbody displacement as defined in §20.3.2. Also, the permissible traction distributions are restricted by the equilibrium conditions (20.41, 20.42). Notice incidentally that the constant C in (20.44) can be wrapped into the arbitrary rigid-body translation ¯ = C on S is ψ = C A0 in (20.31), since a particular solution of the problem ψ(ζ, ζ) (constant) throughout Ω.

338

20 Application to Elasticity Problems

20.6.1 Solution of the interior problem for the circle As in §19.6.1, we can use the Cauchy integral theorem to obtain the solution of the in-plane problem for the region Ω bounded by the circle r = a with prescribed boundary tractions or displacements. Applying the Cauchy integral operator to both sides of equation (20.45), we obtain





1 1 1 θ(s)ds γ χ(s)ds s χ (s)ds f (s)ds + + = . 2πı (s − ζ) 2πı (s − ζ) 2πı (s − ζ) 2πı (s − ζ) (20.46) Since χ is a holomorphic function of ζ in Ω, equation (19.26) gives

1 χ(s)ds = χ(ζ) . 2πı (s − ζ) Also, writing χ=

∞ 

A n ζ n ; χ =

n=0

∞ 

n A¯ n ζ¯ n−1 ,

(20.47)

n=1

and noting that s = a 2 /s on the boundary S, we have s χ (s) =

∞  n A¯ n a 2n−2 n=1

s n−2

.

Using the residue theorem, the second term in (20.46) can then be evaluated as 1 2πı



∞  n A¯ n a 2n−2 s χ (s)ds ds = = A¯ 1 ζ + 2 A¯ 2 a 2 , n−2 (s − ζ) (s − ζ) 2πı s n=1

since all the remaining terms in the series integrate to zero as in (19.24). A similar argument can be applied to the third integral in (20.46). Expanding θ(ζ) =

∞ 

Bk ζ k ,

k=0

we obtain

1 2πı



θ(s)ds = B¯ 0 , (s − ζ)

and we note from §20.3.2 that this can be set to zero without loss of generality. Assembling these results, we conclude that 1 γχ(ζ) = 2πı



f (s)ds − A¯ 1 ζ − 2 A¯ 2 a 2 . (s − ζ)

(20.48)

20.6 Boundary-value Problems

339

To determine the constants A¯ 1 , A¯ 2 , we first replace χ in (20.48) by its Taylor series (20.47) giving 1 2πı



f (s)ds = A¯ 1 ζ + 2 A¯ 2 a 2 + γ A0 + γ A1 ζ + γ A2 ζ 2 + ... . (s − ζ)

Differentiating twice with respect to ζ, we obtain

1 f (s)ds − = A¯ 1 + γ A1 + 2γ A2 ζ + ... 2πı (s − ζ)2

1 f (s)ds = 2γ A2 + 6γ A3 ζ + ... πı (s − ζ)3 and evaluating these last two expressions at ζ = 0,



1 f (s)ds f (s)ds ¯ 1 + γ A1 = − 1 γ A2 = ; A . 2πı s3 2πı s2

(20.49)

Equations (20.49) are sufficient to determine the required constants for the displacement problem (γ = κ), but a degeneracy occurs for the traction problem where γ = 1 and the second equation reduces to

1 ψ(s)ds ¯ . (20.50) A1 + A1 = − 2πı s2 This implies that the right-hand-side must evaluate to a real constant, which (i) places a restriction on the permissible values of ψ(s) and (ii) leaves the imaginary part of A1 indeterminate. We saw already in §20.3.2 that the imaginary part of A1 corresponds to an arbitrary rigid-body rotation. The condition that the right-hand side of (20.50) be real is exactly equivalent to the condition (20.43) enforcing moment equilibrium. To establish the connection, we note that for the circular boundary s=

a2 s

so that

ds = −

a 2 ds . s2

Using this result in (20.43), we obtain  

 

ψ(s)ds 2 =0,  ψds = − a s2 S S showing that the integral is pure imaginary and hence that the right-hand side of (20.50) is real. Once χ is determined, we can apply a similar method to determine θ. We first write the conjugate of equation (20.45) as θ(s) = f (s) − γχ(s) − sχ (s) ; s ∈ S ,

(20.51)

340

20 Application to Elasticity Problems

and apply the Cauchy integral theorem, obtaining





f (s)ds χ(s)ds sχ (s)ds 1 1 γ 1 θ(s)ds θ(ζ) = = − − . 2πı (s − ζ) 2πı (s − ζ) 2πı (s − ζ) 2πı (s − ζ) The last two integrals on the right-hand side can be evaluated using (20.47) and the residue theorem leading to  

a 2 χ (ζ) − χ (0) f (s)ds 1 − γ A¯ 0 − . θ(ζ) = 2πı (s − ζ) ζ Once this expression has been evaluated, the remaining constant A¯ 0 can be chosen to satisfy the condition θ(0) = 0 as discussed in §20.31. However, there is no essential need to do this, since this constant corresponds merely to a rigid-body displacement and makes no contribution to the stress components.

20.6.2 Solution of the exterior problem for the circle For the exterior problem, we again apply the Cauchy integral operator, leading to (20.46), which we repeat here for clarity γ 2πı



1 χ(s)ds + (s − ζ) 2πı



1 s χ (s)ds + (s − ζ) 2πı



1 θ(s)ds = (s − ζ) 2πı



f (s)ds . (s − ζ) (20.52)

In this equation, the first term γ 2πı



χ(s)ds = −γχ(ζ) , (s − ζ)

from (19.30). For the third term, we note that θ(ζ) is holomorphic in the exterior region and hence permits a Laurent expansion θ(ζ) = so θ(s) =

B1 B2 + 2 + ... ζ ζ

B¯ 2 B¯ 1 B¯ 1 s B¯ 2 s 2 + 2 + ... = 2 + 4 + ... . s a a s

This can be regarded as the surface value of a function that is holomorphic in the interior region and hence, since ζ is outside the contour S, 1 2πı



θ(s)ds =0. (s − ζ)

(20.53)

20.6 Boundary-value Problems

341

For the second term in (20.52), we write χ(ζ) =

A1 A2 A3 + 2 + 3 + ... ζ ζ ζ

(20.54)

so χ = −

A¯ 1 2 A¯ 2 3 A¯ 3 − − + ... ; ζ¯2 ζ¯3 ζ¯4

s χ (s) = −

A¯ 1 s 3 2 A¯ 2 s 4 3 A¯ 3 s 5 − − + ... 4 6 a a a8

and as before, we conclude that 1 2πı



sχ (s)ds =0, (s − ζ)

so γχ(ζ) = −

1 2πı



f (s)ds . (s − ζ)

(20.55)

To determine θ, the conjugate equation (20.51) and (19.30) lead to





1 1 γ 1 f (s)ds χ(s)ds sχ (s)ds θ(s)ds θ(ζ)=− =− + + , 2πı (s − ζ) 2πı (s − ζ) 2πı (s − ζ) 2πı (s − ζ) and the second integral on the right-hand side is clearly of the same form as (20.53) and is therefore zero. For the third integral, we have

2  1 sχ (s)ds 1 a χ (s)ds a 2 χ (ζ) = =− , 2πı (s − ζ) 2πı s(s − ζ) ζ using (19.30). We conclude that 1 θ(ζ) = − 2πı



a 2 χ (ζ) f (s)ds − . (s − ζ) ζ

(20.56)

Example To illustrate the procedure, we consider the problem shown in Figure 20.2 in which a circular hole of radius a is loaded by equal and opposite compressive forces F on the diameter. In this case, we have Fδ(θ) Fδ(θ − π) T = − a a and a suitable integral of (20.39) yields ψ(s) = −

ıF sgn(θ) 2

so

ψ(s) =

ıF sgn(θ) , 2

(20.57)

342

20 Application to Elasticity Problems

a F

F

Fig. 20.2 Hole in a large body loaded by equal and opposite forces F.

where sgn(θ) = +1 in 0 < θ < π and −1 in −π < θ < 0. Notice that in performing this integral, it is necessary to traverse the contour in the direction that keeps the material of the body on the left, as explained in §20.5. Substituting in equation (20.55) with γ = 1, f (s) = ψ(s), we obtain 

s=a s=−a  F 1 ψ(s)ds   =− ln(s − ζ) − ln(s − ζ) s=−a s=a 2πı (s − ζ) 4π   ζ +a F ln , = 2π ζ −a

χ(ζ) = −

from which χ (ζ) = −

Fa . − a2)

π(ζ 2

The integral term in (20.56) is evaluated in the same way as 1 2πı



  F ζ +a ψ(s)ds = ln (s − ζ) 2π ζ −a

and it follows that θ(ζ) = −

  ζ +a Fa 3 F ln + . 2π ζ −a πζ(ζ 2 − a 2 )

Finally, the complex stresses can be recovered by substitution into (20.27, 20.28), giving      1 1 Fa  + Θ =2 χ +χ =− π (ζ 2 − a 2 ) (ζ¯2 − a 2 )  2Fa  Φ = −2 ζ χ + θ = π



¯ 2(a 2 − ζ ζ) 1 − 2 2 2 ¯ ¯ (ζ − a ) ζ2

 .

This problem could have been solved using the real stress function approach of Chapter 13 (for example by superposing the solution of Problem 13.2 for two forces at θ = 0, π), but the present method is considerably more direct.

20.7 Conformal Mapping for In-plane Problems

343

20.7 Conformal Mapping for In-plane Problems The technique of conformal mapping introduced in §19.7 can be applied to in-plane problems, but requires some modification because of the presence of the derivative term χ in the boundary condition (20.45). Suppose that the problem is defined in the domain Ω ∗ and that we can identify a mapping function ω(ζ) that maps each point ω ∈ Ω ∗ into a point ζ ∈ Ω, where Ω represents the domain either interior or exterior to the unit circle |ζ| = 1. The boundary condition (20.45) requires that γχ(ω) + ω χ (ω) + θ(ω) = f (ω) ; ω ∈ S ∗ ,

(20.58)

where S ∗ is the boundary of Ω ∗ . If we substitute ω = ω(ζ), the functions χ, θ will become functions of ζ, but to avoid confusion, we shall adopt the notation χζ (ζ) = χ(ω(ζ)) ; θζ (ζ) = θ(ω(ζ)) ;

f ζ (ζ) = f (ω(ζ))

for the corresponding functions in the ζ-plane. We then have χ (ω) ≡

χζ dχ dχζ dζ = =  . dω dζ dω ω

Using the conjugate of this result in (20.58), we obtain γχζ (s) +

ω(s) ω  (s)

χζ (s) + θζ (s) = f ζ (s) ; s ∈ S ,

(20.59)

where S is the unit circle. This defines the boundary-value problem in the ζ-plane. The mathematical arguments used in §§20.6.1, 20.6.2 can be applied to this equation and will yield identical results except in regard to the second term. To fix ideas, suppose Ω ∗ is the region exterior to a non-circular hole and ω(ζ) maps into the region exterior to the unit circle S. Then, the second term in equation (20.52) must be replaced by

ω(s)χζ (s)ds 1 (20.60) 2πı S ω  (s)(s − ζ) and after evaluating the remaining terms as in §20.6.2, we obtain 1 γχζ (ζ) = − 2πı

S

1 f ζ (s)ds + (s − ζ) 2πı



ω(s)χζ (s)ds S

ω  (s)(s − ζ)

.

(20.61)

344

20 Application to Elasticity Problems

The integral (20.60) will be zero if the integrand has no poles inside the contour and can be evaluated using the residue theorem if it has a finite number of poles. As in §20.6.2, we can represent χζ by its Laurent series (20.54) and hence write χζ (s) = − A¯ 1 s 2 − 2 A¯ 2 s 3 − 3 A¯ 3 s 4 + ...,

(20.62)

since on the unit circle ss = 1. Equation (20.62) can be regarded as the surface values of a holomorphic function that has no poles within the circle, but the mapping function ω must be holomorphic in the region Ω exterior to the circle S, and this generally implies that it will have poles in the interior region. Since both Ω, Ω ∗ extend to infinity, it is reasonable to choose a mapping function ω(ζ) such that concentric circles in the ζ-plane map into contours that become more and more circular as we go further away from the hole. This is achieved by the function B2 B3 B1 + 2 + 3 + ... , (20.63) ω(ζ) = cζ + ζ ζ ζ since at large ζ, only the first term remains, which represents simply a linear scaling. The series in (20.63) may be infinite or finite, but in either case ω(ζ) clearly posesses poles at the origin. If the series is finite, the integrand in (20.60) will have at most a finite number of poles and the integral can be evaluated using the residue theorem. It follows from (19.22) that the resulting expression will contain unknowns representing the derivatives of χζ evaluated at the origin, which in view of (20.62) comprises a finite number of the unknown coefficients A¯ 1 , A¯ 2 , .... However, a sufficient number of linear equations for these coefficients can be obtained by constructing the conjugate χζ from (20.61) and equating it or its derivatives to the corresponding values at the origin. This is of course precisely the same procedure that was used to determine the constants A1 , A2 in §20.6.1. This procedure can be generalized to allow ω to be a rational function ω(ζ) =

C1 ζ + C2 ζ 2 + ... + Cn ζ n , D1 ζ + D2 ζ 2 + ... + Dm ζ m

which will posess simple poles at each of the m zeros of the finite polynomial representing the denominator. In this case, the unknowns appearing in the integral (20.60) will be the values of χζ at each of these poles and a set of equations for determining them is obtained by evaluating (20.61) at each such pole3 .

3

For more details of this procedure, the reader is referred to A. H. England, loc.cit, §5.3, I. S. Sokolnikoff, loc.cit, §84.

20.7 Conformal Mapping for In-plane Problems

345

Once χζ has been determined, θζ can be obtained exactly as in §20.6.2. We take the conjugate of (20.59) and apply the Cauchy integral operator obtaining γ 2πı



χζ (s)ds 1 + (s − ζ) 2πı



ω(s)χζ (s)ds

+

ω  (s)(s − ζ)

1 2πı



θζ (s)ds 1 = (s − ζ) 2πı



f ζ (s)ds (s − ζ)

The first integral is zero, as in §20.6.2 and applying (19.30) to the third term, we obtain



ω(s)χζ (s)ds 1 f ζ (s)ds 1 + . (20.64) θζ (ζ) = − 2πı (s − ζ) 2πı ω  (s)(s − ζ) At this stage, all the quantities on the right-hand side of (20.64) are known and the integrals can generally be evaluated using the residue theorem.

20.7.1 The elliptical hole A particularly simple case concerns the elliptical hole for which 

m ω=c ζ+ ζ

 ,

(20.65)

from (19.40), where c, m are real constants. We then have     m ω  (s) = c 1 − 2 = c 1 − ms 2 s and

ω(s) ω  (s)

=

1 s



s2 + m 1 − ms 2

 .

This expression has a simple pole at the origin, but when introduced √into (20.60) this is cancelled by the factor s 2 in (20.62). The simple poles at s = ± 1/m lie outside S, since m < 1. It follows that the integral (20.60) is zero and γχζ (ζ) = −

1 2πı

S

f ζ (s)ds , (s − ζ)

from (20.61). Using (20.65), in (20.64), we obtain θζ (ζ) = −

1 2πı



1 f ζ (s)ds + (s − ζ) 2πı



s(1 + ms 2 )χζ (s)ds (s 2 − m)(s − ζ)

(20.66)

346

20 Application to Elasticity Problems

and since the integrand in the second term has no poles in the region exterior to the unit circle (except for the Cauchy term (s − ζ)), we can evaluate it using the Cauchy integral theorem (19.30), obtaining θζ (ζ) = −

1 2πı



ζ(1 + mζ 2 )χζ (ζ) f ζ (s)ds − . (s − ζ) (ζ 2 − m)

(20.67)

Example: The plane crack opened in tension The ellipse corresponding to the mapping function (20.65) has semi-axes a = c(1 + m) ; b = c(1 − m) , from equation (19.42), so with m = 1, we have a = 2c, b = 0. This defines an ellipse with zero minor axis, which therefore degenerates to a plane crack occupying the line −a < ξ < a, η = 0. Equation (20.65) then takes the special form     a ¯ 1 a 1 ω= ζ+ ; ω= ζ+ . (20.68) 2 ζ 2 ζ¯ We shall use this and the results of §20.7.1 to determine the complete stress field in a cracked body opened by a far-field uniaxial tensile stress σηη = S. This problem was solved using a distribution of dislocations in §13.3.2. As usual, we start by considering the unperturbed solution, which is a state of uniform uniaxial tension σξξ = σξη = 0, σηη = S and hence Θ = S ; Φ = −S , from (20.25, 20.24) respectively. It is clear from equations (20.27, 20.28) that a particular solution can be obtained in which the unperturbed potentials χ0 , θ0 are linear functions of ω. Elementary operations show that the required potentials are χ0 =

Sω Sω ; θ0 = 4 2

and hence

ψ0 =

Sω Sω + , 2 2

from (20.29). For the complete (perturbed) solution, we want the crack S ∗ to be traction free and hence ψ(ω) ≡ ψ0 (ω) + ψ1 (ω) = 0 for ω ∈ S ∗ , where ψ1 is the perturbation due to the crack. Using (20.68) to map the perturbation into the ζ-plane and then setting ζ = s, ζ¯ = 1/s, corresponding to the unit circle |ζ| = 1, we have   1 Sa s+ . ψζ (s) = − 2 s Substituting this result into (20.66) with γ = 1, f ζ = ψζ , we obtain

20.7 Conformal Mapping for In-plane Problems

χζ (ζ) =

1 2πı



Sa 2

347

  1 ds s+ . s (s − ζ)

(20.69)

Since this is an exterior problem, ζ is outside the unit circle and the only pole in the integrand is that at s = 0. The residue theorem therefore gives χζ (ζ) =

Sa Sa 1 =− . 2 (−ζ) 2ζ

(20.70)

Substituting into (20.67) and noting that   Sa 1 Sa + s ; χζ = 2 , ψζ (s) = − 2 s 2ζ we then obtain θζ (ζ) =

1 2πı



Sa 2



1 +s s



ds Sa(ζ 2 + 1) − . (s − ζ) 2ζ(ζ 2 − 1)

The contour integral is identical with that in (20.69) and after routine calculations we find Saζ . (20.71) θζ (ζ) = − 2 (ζ − 1) To move back to the ω-plane, we use (19.44), which with m = 1, c = a/2 takes √ the form ω + ω2 − a2 . ζ= a Substituting this result into (20.70, 20.71), adding in the unperturbed solution χ0 , θ0 and using the identity √ √ 1 ω − ω2 − a2 ω − ω2 − a2 = = 2 , √ ω − (ω 2 − a 2 ) a2 ω + ω2 − a2 we obtain the complete solution to the crack problem as √ S ω2 − a2 Sω Sa 2 Sω + ; θ(ω) = − √ . χ(ω) = − 4 2 2 2 ω2 − a2 Finally, the complex stresses are recovered from (20.25, 20.24) as   Sω Sω Θ = 2 χ + χ = −S + √ + 2 2 ω −a ω2 − a2  Sa 2 (ω − ω) Φ = −2 ωχ + θ = −S + . 2(ω 2 − a 2 )3/2

348

20 Application to Elasticity Problems

Notice that these expressions define the stress field throughout the cracked body; not just the tractions on the plane y = 0 as in the solution in §13.3.2.

Problems 20.1. A state of uniform antiplane shear σx z = S, σ yz = 0 in a large block of material is perturbed by the presence of a rigid circular inclusion whose boundary is defined by the equation r = a in cylindrical polar coördinates r, θ, z. The inclusion is perfectly bonded to the elastic material and is prevented from moving, so that u z = 0 at r = a. Use the complex-variable formulation to find the complete stress field in the block and hence determine the appropriate stress-concentration factor. 20.2. Show that the function (19.40) maps the surfaces of the crack −a < ξ < a, η = 0 onto the unit circle if m = 1, c = a/2. Use this result and the complex-variable formulation to solve the antiplane problem of a uniform stress field σηz = S perturbed by a traction-free crack. In particular, find the mode III stress-intensity factor K III . 20.3. Show that the function ω(ζ) = ζ λ maps the wedge-shaped region 0 < ϑ < α in polar coördinates (ρ, ϑ) into the half-plane y > 0 if λ = α/π. Use this mapping to solve the antiplane problem of the wedge of subtended angle α loaded only by a concentrated out-of-plane force F at the point ρ = a, ϑ = 0. 20.4. Use the stress-transformation equations (1.9–1.11) to prove the relation (20.26). 20.5. Express the plane strain equilibrium equations [(2.1, 2.2) with σx z = σ yz = 0] ¯ in terms of Θ, Φ, p and ζ, ζ. 20.6. By analogy with equations (20.24, 20.25), we can define the complex strains e, ε as e = ex x + e yy ; ε = ex x + 2ıex y − e yy , where we note that e = div u is the dilatation in plane strain. Express the elastic constitutive law (1.54) as a relation between e, ε and Θ, Φ. 20.7. Express the compatibility equation (2.5) in terms of the complex strains e, ε ¯ defined in Problem 20.6 and ζ, ζ. 20.8. Use (20.6) and (1.34) to obtain a complex-variable expression for the inplane rotation ωz . Then apply your result to equation (20.31) to verify that the term involving C1 is consistent with a rigid-body rotation. 20.9. Use the method of §20.4 to determine the displacements due to the Airy stress function φ =r 2 θ. Express the results in terms of the components u r , u θ , using the transformation equation (20.9).

Problems

349

20.10. Substitute the Laurent expansion (20.54) into the contour integral

S

sχ (s)ds , (s − ζ)

where ζ is a point outside the circular contour S of radius a. Use the residue theorem to evaluate each term in the resulting series and hence verify that 1 2πı



a 2 χ (ζ) sχ (s)ds =− . (s − ζ) ζ

20.11. Use the method of §20.6.1 to find the complex stresses Θ, Φ for Problem 12.1, where a disk of radius a is compressed by equal and opposite forces F acting at the points ζ = ±a. 20.12. A state of uniform uniaxial stress, σx x = S is perturbed by the presence of a traction-free circular hole of radius a. Use the method of §20.6.2 to find the complex stresses Θ, Φ. (Note: the Airy stress function solution of this problem is given in §8.4.1.) 20.13. A state of uniform shear stress, σx y = S is perturbed by the presence of a bonded rigid circular inclusion of radius a. Use the method of §20.6.2 to find the complex stresses Θ, Φ. 20.14. A state of uniform uniaxial stress, σ yy = S is perturbed by the presence of a traction-free elliptical hole of semi-axes a, b. Find the complex potentials χ, θ as functions of ω and hence determine the stress-concentration factor as a function of the ratio b/a. Verify that it tends to the value 3 when b = a. 20.15. Use the method of §20.7.1 to find the complete stress field for Problem 13.4 in which a state of uniform shear stress σx y = S is perturbed by the presence of a crack occupying the region −a < x < a, y = 0.

Part V

Three-Dimensional Problems

Chapter 21

Displacement Function Solutions

In Part II, we chose a representation for stress which satisfied the equilibrium equations identically, in which case the compatibility condition leads to a governing equation for the potential function. In three-dimensional elasticity, it is more usual to use the opposite approach — i.e. to define a potential function representation for displacement (which therefore identically satisfies the compatibility condition) and allow the equilibrium condition to define the governing equation. A major reason for this change of method is simplicity. Stress function formulations of three-dimensional problems are generally more cumbersome than their displacement function counterparts, because of the greater complexity of the threedimensional compatibility conditions. It is also worth noting that displacement formulations have a natural advantage in both two and three-dimensional problems when displacement boundary conditions are involved — particularly those associated with multiply-connected bodies (see §2.2.1).

21.1 The Strain Potential The simplest displacement function representation is that in which the displacement u is set equal to the gradient of a scalar function — i.e. 2μu = ∇φ .

(21.1)

The displacement must satisfy the equilibrium equation (2.16), which in the absence of body force reduces to ∇div u + (1 − 2ν)∇ 2 u = 0 .

(21.2)

We therefore obtain ∇∇ 2 φ = 0 , © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 J. R. Barber, Elasticity, Solid Mechanics and Its Applications 172, https://doi.org/10.1007/978-3-031-15214-6_21

353

354

21 Displacement Function Solutions

and hence ∇2φ = C ,

(21.3)

where C is an arbitrary constant. The displacement function φ is generally referred to as the strain or displacement potential. The general solution of equation (21.3) is conveniently considered as the sum of a (harmonic) complementary function and a particular integral, and for the latter, we can take an arbitrary second-order polynomial corresponding to a state of uniform stress. Thus, the representation (21.1) can be decomposed into a uniform state of stress superposed on a solution for which φ is harmonic. It is easy to see that this representation is not sufficiently general to represent all possible displacement fields in an elastic body, since, for example, the rotation  ωz =

∂u y ∂u x − ∂x ∂y

 =

∂2φ ∂2φ − ≡0. ∂x∂ y ∂ y∂x

In other words, the strain potential can only be used to describe irrotational deformation fields1 . It is associated with the name of Lamé (whose name also attaches to the Lamé constants λ, μ), who used it to solve the problems of an elastic cylinder and sphere under axisymmetric loading. In the following work it will chiefly be useful as one term in a more general representation.

21.2 The Galerkin Vector The representation of the previous section is insufficiently general to describe all possible states of deformation of an elastic body, and in seeking a more general form it is natural to examine representations in which the displacement is built up from second rather than first derivatives of a potential function. Using the index notation of §1.1.2, the second derivative term ∂2 F , ∂xi ∂x j where F is a scalar, will represent a Cartesian tensor if the indices i, j are distinct. Alternatively if we make i and j the same, we obtain ∂2 F ∂2 F ∂2 F ∂2 F ≡ + + = ∇2 F , ∂xi ∂xi ∂x1 ∂x1 ∂x2 ∂x2 ∂x3 ∂x3

1

We encountered the same lack of generality with the body force potential of Chapter 7 (see §7.1.1).

21.2 The Galerkin Vector

355

which defines the Laplace operator and returns a scalar. What we want to obtain is a vector, which has one free index, and this can only be done if F also has an index (and hence represents a vector). Then we have a choice of pairing the index on F with one of the derivatives or leaving it free and pairing the two derivatives. The most general form is a linear combination of the two — i.e. 2μu i = C

∂2 Fj ∂ 2 Fi − , ∂x j ∂x j ∂xi ∂x j

(21.4)

where C is an arbitrary constant which we shall later assign so as to simplify the representation. The function Fi must be chosen to satisfy the equilibrium equation (21.2), which in index notation takes the form ∂2uk ∂2ui + (1 − 2ν) =0, ∂xi ∂xk ∂xk ∂xk

(21.5)

from (2.15). Substituting (21.4) into (21.5), and cancelling non-zero factors, we obtain C

∂4 Fj ∂ 4 Fi ∂ 4 Fk − + C(1 − 2ν) ∂x j ∂x j ∂xi ∂xk ∂xk ∂x j ∂xi ∂xk ∂x j ∂x j ∂xk ∂xk −(1 − 2ν)

∂4 Fj =0. ∂xi ∂x j ∂xk ∂xk

Each term in this equation contains one free index i, all the other indices being paired off in an implied sum. In the first, second and fourth term, the index on F is paired with one of the derivatives and the free index appears in one of the derivatives. After expanding the implied summations, these terms are therefore all identical and the representation can be simplified by making them sum to zero. This is achieved by choosing C − 1 − (1 − 2ν) = 0 or C = 2(1 − ν) , after which the representation (21.4) takes the form 2μu i = 2(1 − ν)

∂2 Fj ∂ 2 Fi − , ∂x j ∂x j ∂xi ∂x j

(21.6)

or in vector notation 2μu = 2(1 − ν)∇ 2 F − ∇div F .

(21.7)

With this choice, the equilibrium equation reduces to ∂ 4 Fi =0 ∂x j ∂x j ∂xk ∂xk and the vector function F is biharmonic.

or

∇4 F = 0

(21.8)

356

21 Displacement Function Solutions

This solution is due to Galerkin and the function F is known as the Galerkin vector. A more detailed derivation is given by Westergaard2 , who gives expressions for the stress components in terms of F and who uses this representation to solve a number of classical three-dimensional problems, notably those involving concentrated forces in the infinite or semi-infinite body. We also note that the Galerkin solution was to some extent foreshadowed by Love3 , who introduced a displacement function appropriate for an axisymmetric state of stress in a body of revolution. Love’s displacement function is in fact one component of the Galerkin vector F.

21.3 The Papkovich-Neuber Solution A closely related solution is that developed independently by Papkovich and Neuber in terms of harmonic functions. In general, solutions in terms of harmonic functions are easier to use than those involving biharmonic functions, because the properties of harmonic functions have been extensively studied in the context of other physical theories such as electrostatics, gravitation, heat conduction and fluid mechanics. We define the vector function 1 ψ = − ∇2 F 2

(21.9)

and since the function F is biharmonic, ψ must be harmonic. It can then be verified by differentiation that ∇ 2 (r · ψ) ≡ ∇ 2 (xψx + yψ y + zψz ) ∂ψ y ∂ψz ∂ψx +2 +2 =2 ∂x ∂y ∂z = 2 div ψ .

(21.10)

Substituting for ψ from equation (21.9) into the right-hand side of (21.10), we obtain ∇ 2 (r · ψ) = −div ∇ 2 F = −∇ 2 div F and hence − div F = r · ψ + φ ,

(21.11)

where φ is an arbitrary harmonic function.

2

H. M. Westergaard, Theory of Elasticity and Plasticity, Dover, New York, 1964, §66. A. E. H.Love, A Treatise on the Mathematical Theory of Elasticity, 4th.edn., Dover, New York, 1944, §188.

3

21.3 The Papkovich-Neuber Solution

357

Finally, we substitute (21.9, 21.11) into the Galerkin vector representation (21.7) obtaining 2μu = −4(1 − ν)ψ + ∇(r · ψ + φ) , (21.12) where ψ and φ are both harmonic functions — i.e. ∇2ψ = 0 ; ∇2φ = 0 , but notice that ψ is a vector function, whilst φ is a scalar. This is known as the Papkovich-Neuber solution and it is widely used in modern treatments of threedimensional elasticity problems. We note that the scalar function φ is the harmonic strain potential introduced in §21.1 above.

21.3.1 Change of coördinate system One disadvantage of the Papkovich-Neuber solution is that the term r · ψ introduces a dependence on the origin of coördinates, so that the potentials corresponding to a given displacement field will generally change if the origin is changed. Suppose that a given displacement field is characterized by the potentials ψ 1 , φ1 , so that 2μu = −4(1 − ν)ψ 1 + ∇(r 1 ·ψ 1 + φ1 ) ,

(21.13)

where r 1 is a position vector defining the distance from an origin O1 . We wish to determine the appropriate potentials in a new coördinate system with origin O2 . If the coördinates of O1 relative to O2 are defined by a vector s, the position vector r 2 in the new coördinate system is r2 = s + r1 .

(21.14)

Solving (21.14) for r 1 and substituting in (21.13), we have 2μu = −4(1 − ν)ψ 1 + ∇[(r 2 − s)· ψ 1 + φ1 ] , or 2μu = −4(1 − ν)ψ 2 + ∇(r 2 ·ψ 2 + φ2 ) , where ψ 2 = ψ 1 ; φ2 = φ1 − s · ψ 1 .

(21.15)

Thus, the vector potential ψ is unaffected by the change of origin, but there is a change in the scalar potential φ. By contrast, the Galerkin vector F is independent of the choice of origin.

358

21 Displacement Function Solutions

21.4 Completeness and Uniqueness It is a fairly straightforward matter to prove that the representations (21.6, 21.12) are complete — i.e. that they are capable of describing all possible elastic displacement fields in a three dimensional body4 . The related problem of uniqueness presents greater difficulties. Both representations are redundant, in the sense that there are infinitely many combinations of functions which correspond to a given displacement field. In a non-rigorous sense, this is clear from the fact that a typical elasticity problem will be reduced to a boundary-value problem with three conditions at each point of the boundary (three tractions or three displacements or some combination of the two). We should normally expect such conditions to be sufficient to determine three harmonic potential functions in the interior, but (21.12) essentially contains four independent functions (three components of ψ and φ). Neuber gave a ‘proof’ that one of these four functions can always be set equal to zero, but it has since been found that his argument only applies under certain restrictions on the shape of the body. A related ‘proof’ for the Galerkin representation is given by Westergaard in his §70. We shall give a summary of one such argument and show why it breaks down if the shape of the body does not meet certain conditions. We suppose that a solution to a certain problem is known and that it involves all the three components ψx , ψ y , ψz of ψ and φ. We wish to find another representation of the same field in which one of the components is everywhere zero. We could achieve this if we could construct a ‘null’ field — i.e. one which corresponds to zero displacement and stress at all points — for which the appropriate component is the same as in the original solution. The difference between the two solutions would then give the same displacements as the original solution (since we have subtracted a null field only) and the appropriate component will have been reduced to zero. Equation (21.12) shows that ψ, φ will define a null field if ∇(r · ψ + φ) − 4(1 − ν)ψ = 0 .

(21.16)

Taking the curl of this equation, we obtain curl ∇(r · ψ + φ) − 4(1 − ν)curl ψ = 0 and hence curl ψ = 0 ,

(21.17)

since curl ∇ of any function is identically zero. We also note that by applying the operator div to equation (21.16), we obtain

4

See for example, H. M. Westergaard, loc. cit. §69 and R. D. Mindlin (1936), Note on the Galerkin and Papcovich stress functions, Bulletin of the American Mathematical Society, Vol. 42, pp. 373– 376. See also Problem 21.10.

21.4 Completeness and Uniqueness

359

∇ 2 (r · ψ + φ) − 4(1 − ν)div ψ = 0 and hence, substituting from equation (21.10) for the first term, 2div ψ − 4(1 − ν)div ψ = −2(1 − 2ν)div ψ = 0 .

(21.18)

Now equation (21.17) is precisely the condition which must be satisfied for the function ψ to be expressible as the gradient of a scalar function5 . Hence, we conclude that there exists a function H such that ψ = ∇H .

(21.19)

Furthermore, substituting (21.19) into (21.18), we see that H must be a harmonic function. In order to dispense with one of the components (say ψz ) of ψ in the given solution, we therefore need to be able to construct a harmonic function H to satisfy the equation ∂H = ψz . (21.20) ∂z If we can do this, it is easily verified that the appropriate null field can be found, since on substituting into equation (21.16) we get ∇(r·∇ H + φ − 4(1 − ν)H ) = 0 , which can be satisfied by choosing φ so that φ = 4(1 − ν)H − r·∇ H . Noting that   ∂H ∂H ∂H +y +z ∇ 2 (r·∇ H ) = ∇ 2 x ∂x ∂y ∂z = 2∇ 2 H , we see that φ will then be harmonic, as required, since H is harmonic.

21.4.1 Methods of partial integration We therefore consider the problem of integrating equation (21.20) subject to the constraint that the integral be a harmonic function. We shall examine two methods of doing this which give some insight into the problem. 5

See §7.1.1.

360

21 Displacement Function Solutions

The given solution and hence the function ψz is only defined in the finite region occupied by the elastic body which we denote by Ω. We also denote the boundary of Ω (not necessarily connected) by Γ . One candidate for the required function H is obtained from the integral  H1 =

z

ψz dz ,

(21.21)

zΓ

where the integral is performed from a point on Γ to a point in the body along a line in the z-direction. This function is readily constructed and it clearly satisfies equation (21.20), but it will not in general be harmonic. However, ∂ H1 ∂ 2 ∇ H1 = ∇ 2 =0 ∂z ∂z

(21.22)

and hence ∇ 2 H1 = f (x, y) . Now a harmonic function H satisfying equation (21.20) can be constructed as H = H1 + H2 , where H2 is any function of x, y only that satisfies the equation ∇ 2 H2 = − f (x, y) .

(21.23)

A suitable choice for H2 is the logarithmic potential due to the source distribution f (x, y), which can be explicitly constructed as a convolution integral on the logarithmic (two-dimensional) point source solution. Notice that H2 is not uniquely determined by (21.23), but we are looking for a particular integral only, not a general solution. Indeed, this freedom of choice of solution could prove useful in extending the range of conditions under which the integration of (21.20) can be performed. At first sight, the procedure described above seems to give a general method of constructing the required function H , but a difficulty is encountered with the definition of the lower limit z Γ in the integral (21.21). The path of integration is along a straight line in the z-direction from Γ to the general point (x, y, z). There may exist points for which such a line cuts the surface Γ at more than two points — i.e. for which z Γ is multivalued. Such a case is illustrated as the point P(x, y, z) in Figure 21.1. The integrand is undefined outside Ω and hence we can only give a meaning to the definition (21.21) if we choose the point A as the lower limit. However, by the same token, we must choose the point C as the lower limit in the integral for the point Q(x, y, z  ). It follows that the function f (x, y) is not truly a function of x, y only, since it has different values at the points P, Q, which have the same coördinates (x, y). The f (x, y) appropriate to the region above the line T T  will generally have a

21.4 Completeness and Uniqueness

361

Fig. 21.1 Path of integration for equation (21.21).

discontinuity across the line T R, where T is the point of tangency of Γ to lines in the z-direction (see Figure 21.1). Notice, however that the freedom of choice of the integral of equation (21.23) may in certain circumstances permit us to eliminate this discontinuity. Another way of developing the function H1 is as follows:- We first find that distribution of sources which when placed on the surface Γ in an infinite space would give the required potential ψz in Ω. This distribution is unique and gives a form of continuation of ψz into the rest of the infinite space which is everywhere harmonic except on the surface Γ . We then write −∞ for the lower limit in equation (21.21). The function defined by equation (21.21) will now be harmonic except when the path of integration passes through Γ , in which case ∇ 2 H1 will contain a term proportional to the strength of the singularity at the appropriate point(s) of intersection. As these points will be the same for any given value of (x, y), the resulting function will satisfy equation (21.22), but we note that again there will be different values for the function f (x, y) if the appropriate path of integration cuts Γ in more than two points. Furthermore, this method of construction shows that the difference between the two values is proportional to the strength of the singularities on the extra two intersections with Γ (i.e. at points B, C in Figure 21.1). One of the consequences is that f (x, y) in the upper region will then show a square root singular discontinuity to the left of the line T R, assuming that Γ has no sharp corners. This question is still a matter of active interest in elasticity6 . 6

For further discussion of the question, see R. A. Eubanks and E. Sternberg (1956), On the completeness of the Boussinesq-Papcovich stress functions, Journal of Rational Mechanics and Analysis, Vol. 5, pp. 735–746; P. M. Naghdi and C. S. Hsu (1961), On a representation of displacements in linear elasticity in terms of three stress functions, Journal of Mathematics and Mechanics, Vol. 10, pp. 233–246; T. Tran Cong and G. P. Steven (1979), On the representation of elastic displacement fields in terms of three harmonic functions, Journal of Elasticity, Vol. 9, pp. 325–333.

362

21 Displacement Function Solutions

21.5 Body Forces So far we have concentrated on the development of solutions to the equation of equilibrium (21.2) without body forces. As in Chapter 7, problems involving body forces are conveniently treated by first seeking a particular solution satisfying the equilibrium equation (2.16) ∇div u + (1 − 2ν)∇ 2 u +

(1 − 2ν) p =0 μ

(21.24)

and then superposing appropriate homogeneous solutions — i.e. solutions without body force — to satisfy the boundary conditions.

21.5.1 Conservative body force fields If the body force p is conservative, we can write p = −∇V

(21.25)

and a particular solution can always be found using the strain potential (21.1). Substituting (21.1, 21.25) into (21.24) we obtain (1 − ν)∇∇ 2 φ − (1 − 2ν)∇V = 0 , which is satisfied if ∇2φ =

(1 − 2ν)V . (1 − ν)

(21.26)

Solutions of this equation can be found for all functions V . The corresponding stress components can then be obtained from the stress-strain and strain-displacement relations. For example, we have σx x = λdiv u + 2μ

λ 2 ∂2φ ∂u x ∂2φ νV = ∇ φ+ + = , ∂x 2μ ∂x 2 (1 − ν) ∂x 2

(21.27)

using (1.51–1.53, 21.26). For the shear stresses, we have for example  σx y = 2μex y = μ

∂u y ∂u x + ∂x ∂y

 =

∂2φ . ∂x∂ y

(21.28)

Problems

363

21.5.2 Non-conservative body force fields If the body force p is non-conservative, a particular solution can always be obtained using the Galerkin vector representation of equation (21.7). Substituting this into (2.15), we obtain ∂ 4 Fi pi , =− ∂x j ∂x j ∂xk ∂xk (1 − ν) or ∇4 F = −

p . (1 − ν)

(21.29)

This constitutes three uncoupled equations relating each component of the Galerkin vector to the corresponding component of the body force. Particular solutions can be found for all functions p.

Problems 21.1. Show that the function ψ=y

∂φ ∂φ −x ∂x ∂y

will be harmonic (∇ 2 ψ = 0) if φ is harmonic. Show also that ψ will be biharmonic if φ is biharmonic. 21.2. Find the most general solution of equation (21.3) that is spherically symmetric  — i.e. depends only on the distance R = x 2 + y 2 +z 2 from the origin. Use your solution to find the stress and displacement field in a hollow spherical container of inner radius b and outer radius a, loaded only by internal pressure p0 at R = b. Note: Strain-displacement relations and expressions for the gradient and Laplacian operator in spherical polar coördinates are given in §22.4.2 below. 21.3. By expressing the vector operators in index notation, show that ∇div ∇div V ≡ ∇ 2 ∇div V ≡ ∇div ∇ 2 V , where V is any vector field. 21.4. Starting from equation (21.6), use (1.37, 1.54, 1.52) to develop general expressions for the stress components σi j in the Galerkin formulation, in terms of Fi and the elastic constants μ, ν.  21.5. Verify that the function ψ = 1/ x 2 + y 2 +z 2 is harmonic everywhere except at the origin, where it is singular. This function can therefore be used as a Papkovich-

364

21 Displacement Function Solutions

Neuber displacement function for a body that does not contain the origin, such as the infinite body with a spherical hole centred on the origin. Show by partial integration that it is impossible to find a function H such that ∂H =ψ ∂z and ∇ 2 H = 0, except possibly at the origin. 21.6. Repeat the derivation of §21.3 using equation (21.29) in place of (21.8) in order to generalize the Papkovich-Neuber solution to include a body force field p. In particular, find the governing equations for the functions ψ, φ. 21.7. Find a particular solution for the Galerkin vector F for the non-conservative body force field px = −ρΩ˙ y ; p y = ρΩ˙ x ; pz = 0 . 21.8. Prove the result (used in deriving (21.17)) that curl ∇φ = 0 , for any scalar function φ. 21.9. (i) Verify that the Galerkin vector F = C x 3 yk satisfies the governing equation (21.8) and determine the corresponding displacement vector u. (ii) Show that   F 0 = (1 − 2ν)∇ 2 + ∇div f defines a null Galerkin vector (i.e. the corresponding displacements are everywhere zero) if ∇ 4 f = 0. (iii) Using this result or otherwise, find an alternative Galerkin vector for the displacement field of part (i) that is confined to the x y-plane — i.e. such that Fz = 0. 21.10. By writing the equilibrium equation (21.2) as ∇2u = −

1 ∇div u , (1 − 2ν)

show that the displacement u can be expressed in the form μu = ∇χ + Ψ , where Ψ is a harmonic vector potential and χ is a scalar potential. Find the equation that must be satisfied by χ and show that the resulting representation of displacement is equivalent to the Papkovich-Neuber solution (21.12). Hence or otherwise, show that this representation is complete.

Chapter 22

The Boussinesq Potentials

The Galerkin and Papkovich-Neuber solutions have the advantage of presenting a general solution to the problem of elasticity in a suitably compact notation, but they are not always the most convenient starting point for the solution of particular three-dimensional problems. If the problem has a plane of symmetry or particularly simple boundary conditions, it is often possible to develop a special solution of sufficient generality in one or two harmonic functions, which may or may not be components or linear combinations of components of the Papkovich-Neuber solution. For this reason, it is convenient to record detailed expressions for the displacement and stress components arising from the several terms separately and from certain related displacement potentials. The first catalogue of solutions of this kind was compiled by Boussinesq1 and is reproduced by Green and Zerna2 , whose terminology we use here. Boussinesq identified three categories of harmonic potential, one being the strain potential of §21.1, already introduced by Lamé, and another comprising a set of three scalar functions equivalent to the three components of the Papkovich-Neuber vector, ψ. The third category comprises three solutions particularly suited to torsional deformations about the three axes respectively. In view of the completeness of the PapkovichNeuber solution, it is clear that these latter functions must be derivable from equation (21.12) and we shall show how this can be done in §22.3 below3 .

Supplementary Information The online version contains supplementary material available at https://doi.org/10.1007/978-3-031-15214-6_22. 1

J. Boussinesq, Application des potentiels à l’étude de l’équilibre et du mouvement des solides élastiques, Gauthier-Villars, Paris, 1885. 2 A. E. Green and W. Zerna, Theoretical Elasticity, 2nd.edn., Clarendon Press, Oxford, 1968, pp. 165–176. 3 Other relations between the Boussinesq potentials are demonstrated by J. P. Bentham (1979), Note on the Boussinesq-Papkovich stress functions, J.Elasticity, Vol. 9, pp. 201–206. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 J. R. Barber, Elasticity, Solid Mechanics and Its Applications 172, https://doi.org/10.1007/978-3-031-15214-6_22

365

366

22 The Boussinesq Potentials

22.1 Solution A: The Strain Potential Green and Zerna’s Solution A is identical with the strain potential defined in §21.1 above (with the restriction that ∇ 2 φ = 0) and with the scalar potential φ of the Papkovich-Neuber solution [equation (21.12)]. We define ∂φ ∂φ ∂φ ; 2μu y = ; 2μu z = , 2μu x = ∂x ∂y ∂z and the dilatation

2μe = 2μdiv u = ∇ 2 φ = 0 .

Substituting in the stress-strain relations, we have σx x = λe + 2μex x = and similarly σ yy =

∂2φ ∂x 2

∂2φ ∂2φ ; σ = . zz ∂ y2 ∂z 2

We also have σx y = 2μex y =

∂2φ ∂2φ ∂2φ ; σ yz = ; σzx = . ∂x∂ y ∂ y∂z ∂z∂x

22.2 Solution B Green and Zerna’s Solution B is obtained by setting to zero all the functions in the Papkovich-Neuber solution except the z-component of ψ, which we shall denote by ω. In other words, in equation (21.12), we take ψ = kω ; φ = 0 .

It follows that ∂ω ∂ω ∂ω ; 2μu y = z ; 2μu z = z − (3 − 4ν)ω ∂x ∂y ∂z ∂ω ∂ω ∂ω 2μe = z∇ 2 ω + − (3 − 4ν) = −2(1 − 2ν) , ∂z ∂z ∂z

2μu x = z

since ∇ 2 ω = 0. Substituting these expressions into the stress-strain relations and noting that

22.3 Solution E: Rotational Deformation

367

λ=

2μν , (1 − 2ν)

we find ∂2ω ∂ω ∂2ω ∂ω − 2ν = z − 2ν ; σ yy ∂x 2 ∂z ∂ y2 ∂z 2 2 ∂ ω ∂ω ∂ ω ; σx y = z z 2 − 2(1 − ν) ∂z ∂z ∂x∂ y   1 ∂ω ∂2ω ∂2ω ∂ω +z +z − (3 − 4ν) 2 ∂y ∂ y∂z ∂ y∂z ∂y 2 ∂ω ∂ ω − (1 − 2ν) z ∂ y∂z ∂y ∂2ω ∂ω z − (1 − 2ν) . ∂z∂x ∂x

σx x = λe + 2μex x = z σzz = σ yz = = σzx =

This solution — if combined with solution A — gives a general solution for the torsionless axisymmetric deformation of a body of revolution. The earliest solution of this problem is due to Love4 and leads to a single biharmonic function which is actually the component Fz of the Galerkin vector (§21.2). Solution B is also particularly suitable for problems in which the plane surface z = 0 is a boundary, since the expressions for stresses and displacements take simple forms on this surface. It is clearly possible to write down similar solutions corresponding to PapkovichNeuber vectors in the x- and y-directions by permuting suffices. Green and Zerna refer to these as Solutions C and D.

22.3 Solution E: Rotational Deformation Green and Zerna’s Solution E can be derived (albeit somewhat indirectly) from the Papkovich-Neuber solution by taking the vector function ψ to be the curl of a vector field oriented in the z-direction — i.e. ψ = curl kχ . It follows that r ·ψ = x and hence ∇ 2 (r · ψ) = 2

∂χ ∂χ −y ∂y ∂x

∂2χ ∂2χ −2 =0, ∂x∂ y ∂ y∂x

— i.e. r · ψ is harmonic. We can therefore choose φ such that 4

A. E. H.Love, loc. cit.

368

22 The Boussinesq Potentials

r ·ψ+φ≡0 and hence 2μu = −4(1 − ν)curl kχ . To avoid unnecessary multiplying constants, it is convenient to write Ψ = −2(1 − ν)χ , in which case 2μu x = 2

∂Ψ ∂Ψ ; 2μu y = −2 ; 2μu z = 0 , ∂y ∂x

giving

2μe = 0 .

Substituting in the stress-strain relations, we find ∂2Ψ ∂2Ψ σx x = 2 ; σ yy = −2 ; σzz = 0 ∂x∂ y ∂x∂ y   2 ∂ Ψ ∂2Ψ ∂2Ψ ∂2Ψ ; σ . ; σ σx y = − = − = yz zx ∂ y2 ∂x 2 ∂x∂z ∂ y∂z As with Solution B, two additional solutions of this type can be constructed by permuting suffices.

22.4 Other Coördinate Systems The above results have been developed in the Cartesian coördinates x, y, z, but similar expressions are easily obtained in other coördinate systems.

22.4.1 Cylindrical polar coördinates The cylindrical polar coördinate system r, θ, z is related to the Cartesian system through the equations x = r cos θ ; y = r sin θ as in the two-dimensional transformation (8.1). The gradient operator takes the form ∇ ≡ er

∂ eθ ∂ ∂ + +k , ∂r r ∂θ ∂z

22.4 Other Coördinate Systems

369

where er , eθ are unit vectors in the r and θ directions respectively, and the twodimensional Laplacian operator (8.10) aquires a z-derivative term, becoming ∇2 ≡

∂2 1 ∂ ∂2 1 ∂2 + + . + ∂r 2 r ∂r r 2 ∂θ2 ∂z 2

(22.1)

The in-plane strain-displacement relations (8.12) need to be supplemented by er z =

1 2



∂u r ∂u z + ∂z ∂r

 ; eθz =

1 2



∂u θ 1 ∂u z + ∂z r ∂θ

 ; ezz =

∂u z . ∂z

Calculation of the stress and displacement components then proceeds as in §§22.1–22.3. The corresponding expressions for both Cartesian and cylindrical polar coördinates are tabulated in Table 22.1. Also, these expressions are listed in the Maple and Mathematica files ‘ABExyz’ and ‘ABErtz’.

22.4.2 Spherical polar coördinates The spherical polar coördinate system R, θ, β is related to the cylindrical system by the further change of variables r = R sin β ; where R=

z = R cos β ,

  r 2 + z2 = x 2 + y2 + z2

represents the distance from the origin and the two spherical angles θ, β are equivalent to measures of longitude and lattitude respectively, defining position on a sphere of radius R. Notice however that it is conventional to measure the angle of lattitude from the pole (the positive z-axis) rather than from the equator. Thus, the two poles are defined by β = 0, π and the equator by β = π/2. The gradient operator takes the form ∇ ≡ eR

∂ ∂ eθ eβ ∂ + + , ∂R R sin β ∂θ R ∂β

where er , eθ , eβ are unit vectors in the R, θ and β directions respectively. Transformation of the Laplacian operator, the strain-displacement relations and the expressions for displacement and stress components in Solutions A,B and E is algebraically tedious but mathematically routine and only the principal results are summarized here.

370

22 The Boussinesq Potentials

Table 22.1 Green and Zerna’s Solutions A, B and E Solution A Solution B ∂φ ∂x ∂φ ∂y ∂φ ∂z

2μu x 2μu y 2μu z

σx y σ yy

∂2φ ∂x∂z

σx z

∂2φ ∂ y∂z ∂2φ ∂z 2

σ yz σzz

σr z σθz

∂2ω ∂ω − 2ν ∂x 2 ∂z ∂2ω z ∂x∂ y ∂2ω ∂ω z 2 − 2ν ∂y ∂z

∂2Ψ ∂x∂ y ∂2Ψ ∂2Ψ − 2 ∂y ∂x 2 2 ∂ Ψ −2 ∂x∂ y

∂2ω ∂ω − (1 − 2ν) ∂x∂z ∂x

∂2Ψ ∂ y∂z

z

z

0

∂2ω ∂ω − (1 − 2ν) ∂ y∂z ∂y ∂2ω ∂ω z 2 − 2(1 − ν) ∂z ∂z

z

2

−

∂2Ψ ∂x∂z 0

∂ω ∂r z ∂ω r ∂θ

2 ∂Ψ r ∂θ ∂Ψ −2 ∂r

∂2φ ∂r 2 2 1 ∂ φ 1 ∂φ − 2 r ∂r ∂θ r ∂θ 1 ∂φ 1 ∂2φ + 2 2 r ∂r r ∂θ ∂2φ ∂r ∂z 1 ∂2φ r ∂θ∂z

∂2ω ∂ω − 2ν ∂r 2 ∂z z ∂2ω z ∂ω − 2 r ∂r ∂θ r ∂θ z ∂ω ∂ω z ∂2ω − 2ν + 2 r ∂r r ∂θ2 ∂z ∂2ω ∂ω z − (1 − 2ν) ∂r ∂z ∂r 2 z ∂ ω 1 − 2ν ∂ω − r ∂θ∂z r ∂θ

2 ∂2Ψ 2 ∂Ψ − 2 r ∂r ∂θ r ∂θ 1 ∂2Ψ 1 ∂Ψ ∂2Ψ + 2 − 2 r ∂r ∂r r ∂θ2 2 ∂2Ψ 2 ∂Ψ − + 2 r ∂r ∂θ r ∂θ 1 ∂2Ψ r ∂θ∂z ∂2Ψ − ∂r ∂z

2μu θ

σθθ

∂ω − (3 − 4ν)ω ∂z

∂Ψ ∂y ∂Ψ −2 ∂x 2

∂φ ∂r 1 ∂φ r ∂θ

2μu r

σr θ

z

∂2φ ∂x 2 ∂2φ ∂x∂ y ∂2φ ∂ y2

σx x

σrr

∂ω ∂x ∂ω z ∂y z

Solution E

z

z

The Laplacian operator is ∇2 ≡

1 ∂2 ∂2 2 ∂ 1 ∂2 cot β ∂ + 2 2 + + 2 2+ 2 2 ∂R R ∂R R ∂β R 2 ∂β R sin β ∂θ

(22.2)

22.5 Solutions Obtained by Superposition

371

and the strain-displacement relations5 are u β cot β 1 ∂u θ ∂u R uR ; eθθ = + + ∂R R R R sin β ∂θ   1 ∂u β 1 ∂u R ∂u θ uθ uR 1 + ; e Rθ = + − = R R ∂β 2 R sin β ∂θ ∂R R   u θ cot β 1 ∂u β 1 1 ∂u θ − + = 2 R ∂β R R sin β ∂θ   ∂u β uβ 1 1 ∂u R + − = . 2 R ∂β ∂R R

eR R = eββ eθβ eβ R

The corresponding displacement and stress components in Solutions A,B and E are given in Table 22.2 and are also listed in the Maple and Mathematica files ‘ABErtb’.

22.5 Solutions Obtained by Superposition The solutions obtained in the above sections can be superposed as required to provide a solution appropriate to the particular problem at hand. A case of particular interest is that in which the plane z = 0 is one of the boundaries of the body — for example the semi-infinite solid or half-space z > 0. We consider here two special cases.

22.5.1 Solution F: Frictionless isothermal contact problems If the half-space is indented by a frictionless punch, so that the surface z = 0 is subjected to normal tractions only, a simple formulation can be obtained by combining solutions A and B and defining a relationship between φ and ω in order to satisfy identically the condition σzx = σzy = 0 on z = 0. We write ∂ϕ (22.3) φ = (1 − 2ν)ϕ ; ω = ∂z in solutions A, B respectively, obtaining σzx = z

5

∂3ϕ ∂3ϕ ; σzy = z , 2 ∂x∂z ∂ y∂z 2

See for example A. S. Saada, Elasticity, Pergamon Press, New York, 1973, §6.7.

372

22 The Boussinesq Potentials

Table 22.2 Solutions A, B and E in spherical polar coördinates Solution A Solution B 2μu R

∂φ ∂R

2μu θ

1 ∂φ R sin β ∂θ

2μu β

1 ∂φ R ∂β

cos β

σR R

∂2φ ∂ R2

R cos β

1 ∂φ cot β ∂φ + 2 R ∂R R ∂β +

∂2φ 1 2 R 2 sin β ∂θ2

cot β

∂ω ∂θ

−2

∂Ψ 2 cos β ∂Ψ sin β − ∂R R ∂β 2 cot β ∂Ψ R ∂θ

∂ω + (3 − 4ν)ω sin β ∂β

2 ∂2Ψ 2 ∂Ψ − 2 R ∂ R∂θ R ∂θ

∂2ω ∂ω −2(1 − ν) cos β ∂ R2 ∂R 2ν ∂ω sin β R ∂β

∂ω 2ν ∂ω cos β + sin β ∂R R ∂β

(1 − 2ν) +

2 ∂Ψ R ∂θ

∂ω −(3 − 4ν)ω cos β ∂R

R cos β

+

σθθ

Solution E

2 R 2 sin2

cos2 β ∂ω cot β ∂ 2 ω + R sin β ∂β R sin β ∂θ2

−

∂Ψ 2 ∂2Ψ − ∂θ R ∂ R∂θ β 2 cot β ∂ 2 Ψ R 2 ∂β∂θ

σββ

1 ∂φ 1 ∂2φ + 2 R ∂R R ∂β 2

cos β ∂ 2 ω ∂ω + (1 − 2ν) cos β R ∂β 2 ∂R 2(1 − ν) sin β ∂ω + R ∂β

2 cot β ∂ 2 Ψ 2 cot 2 β ∂Ψ − R 2 ∂θ∂β R2 ∂θ

σθβ

∂2φ 1 R 2 sin β ∂θ∂β

cot β ∂ 2 ω 2(1 − ν) ∂ω + R ∂θ∂β R ∂θ

cot β ∂ 2 Ψ sin β ∂ 2 Ψ − R 2 sin β ∂θ2 R ∂ R∂β

−

cot β ∂φ R 2 sin β ∂θ

σβ R

1 ∂2φ 1 ∂φ − R ∂β∂ R R 2 ∂β

σ Rθ

1 ∂2φ R sin β ∂ R∂θ −

1 R 2 sin β

∂φ ∂θ

−

∂ω 1 R sin2 β ∂θ

−

∂Ψ cos β ∂ 2 Ψ 1 + R 2 ∂β 2 R 2 sin β ∂β

∂ω ∂2ω sin β +cos β ∂R ∂β∂ R 2(1 − ν) cos β ∂ω − R ∂β

1 ∂2Ψ cot β ∂ 2 Ψ + 2 R ∂θ∂β R ∂θ∂ R 2 cot β ∂Ψ − R 2 ∂θ

∂2ω ∂ R∂θ

∂2Ψ 1 ∂2Ψ −sin β R 2 sin β ∂θ2 ∂ R2

(1 − 2ν)

cot β −

2(1 − ν) cot β ∂ω R ∂θ

−

cos β ∂ 2 Ψ 2 cos β ∂Ψ + R ∂ R∂β R 2 ∂β sin β ∂Ψ + R ∂R

which vanish as required on the plane z = 0. The remaining stress and displacement components are listed in Table 22.3 and at this stage we merely note that the normal traction and normal displacement at the surface reduce to the simple expressions

22.5 Solutions Obtained by Superposition

373

Table 22.3 Solutions F and G Solution F

Solution G

Frictionless isothermal 2μu x

z

∂ϕ ∂2ϕ + (1 − 2ν) ∂x∂z ∂x

2(1 − ν)

∂2χ ∂χ +z ∂x ∂x∂z

2μu y

z

∂ϕ ∂2ϕ + (1 − 2ν) ∂ y∂z ∂y

2(1 − ν)

∂2χ ∂χ +z ∂y ∂ y∂z

2μu z

z

∂2ϕ ∂ϕ − 2(1 − ν) ∂z ∂z 2

−(1 − 2ν)

σx x

z

σx y

z

σ yy

z

∂3ϕ ∂2ϕ ∂2ϕ + 2ν 2 + ∂x 2 ∂z ∂x 2 ∂y

∂2ϕ ∂3ϕ + (1 − 2ν) ∂x∂ y∂z ∂x∂ y ∂3ϕ ∂2ϕ ∂2ϕ + 2ν 2 + 2 2 ∂ y ∂z ∂y ∂x

2(1 − ν)

∂2χ ∂3χ ∂2χ + z 2 − 2ν 2 ∂x 2 ∂x ∂z ∂z

2(1 − ν) 2(1 − ν)

∂χ ∂2χ +z 2 ∂z ∂z

∂3χ ∂2χ +z ∂x∂ y ∂x∂ y∂z

∂2χ ∂3χ ∂2χ + z 2 − 2ν 2 2 ∂y ∂ y ∂z ∂z

σx z

z

∂3ϕ ∂x∂z 2

∂2χ ∂3χ +z ∂x∂z ∂x∂z 2

σ yz

z

∂3ϕ ∂ y∂z 2

∂3χ ∂2χ +z ∂ y∂z ∂ y∂z 2

σzz

2μu r 2μu θ σrr σr θ

σθθ

z

z

∂3ϕ ∂2ϕ − ∂z 3 ∂z 2

∂ϕ ∂2ϕ + (1 − 2ν) ∂r ∂z ∂r

z ∂2ϕ (1 − 2ν) ∂ϕ + r ∂θ∂z r ∂θ   3 2 2 ∂ ϕ ∂ ϕ ∂2ϕ ∂ ϕ z 2 − 2ν + + ∂r ∂z ∂r 2 ∂r 2 ∂z 2 z ∂2ϕ z ∂3ϕ − 2 r ∂r ∂θ∂z r ∂θ∂z   (1 − 2ν) ∂ 2 ϕ 1 ∂ϕ + − r ∂r ∂θ r ∂θ −(1 − 2ν) −z

∂2ϕ ∂2ϕ − 2 ∂r ∂z 2

∂3ϕ ∂3ϕ −z 3 2 ∂r ∂z ∂z

z 2(1 − ν)

∂3χ ∂z 3

∂2χ ∂χ +z ∂r ∂r ∂z

2(1 − ν) ∂χ z ∂2χ + r ∂θ r ∂θ∂z ∂2χ ∂3χ ∂2χ + z 2 − 2ν 2 2 ∂r ∂r ∂z ∂z   1 ∂χ 2(1 − ν) ∂ 2 χ − r ∂r ∂θ r ∂θ

2(1 − ν)

+

z ∂3χ z ∂2χ − 2 r ∂z∂r ∂θ r ∂z∂θ

−2(1 − ν) +

∂2χ ∂2χ −2 2 2 ∂r ∂z

z ∂3χ z ∂2χ + 2 r ∂z∂r r ∂z∂θ2

σr z

z

∂3ϕ ∂r ∂z 2

∂2χ ∂3χ +z ∂r ∂z ∂r ∂z 2

σθz

z ∂3ϕ r ∂θ∂z 2

1 ∂2χ z ∂3χ + r ∂θ∂z r ∂θ∂z 2

374

22 The Boussinesq Potentials

σzz = −

∂2ϕ ∂ϕ ; 2μu z = −2(1 − ν) . 2 ∂z ∂z

It follows that frictionless indentation problems for the half-space can be reduced to classical boundary-value problems in the harmonic potential function ϕ. Various examples are considered by Green and Zerna6 , who also show that symmetric problems of the plane crack in an isotropic body can be treated in the same way. These problems are considered in Chapters 31–33 below.

22.5.2 Solution G: The surface free of normal traction The counterpart of the preceding solution is that obtained by combining Solutions A,B and requiring that the normal tractions vanish on the surface z = 0. We shall find this solution useful in problems involving a plane interface between two dissimilar materials (Chapter 35) and also in certain axisymmetric crack and contact problems. The normal traction on the plane z = 0 is σzz =

∂2φ ∂ω , − 2(1 − ν) ∂z 2 ∂z

from Table 22.1 and hence it can be made to vanish by writing φ = 2(1 − ν)χ ; ω =

∂χ , ∂z

where χ is a new harmonic potential function. The resulting stress and displacement components are given as Solution G in Table 22.3.

22.5.3 A plane strain solution If we combine Green and Zerna’s Solutions A and D7 and require that the corresponding harmonic functions be independent of z, we obtain the displacement and stress components

6

A. E. Green and W. Zerna, loc. cit. §§5.8-5.10. Solution D can be obtained from Solution B of Table 22.1 by permuting suffices — i.e. making the substitutions z → y, y → x, x → z in the expressions in Cartesian coördinates.

7

22.6 A Three-dimensional Complex-Variable Solution

2μu x = σx x =

375

∂ω ∂ω ∂φ ∂φ +y ; 2μu y = +y − (3 − 4ν)ω ; 2μu z = 0 ∂x ∂x ∂y ∂y

∂2φ ∂2ω ∂ω ∂2φ ∂2ω ∂ω ; σ + y − 2ν = + y − 2(1 − ν) yy ∂x 2 ∂x 2 ∂y ∂ y2 ∂ y2 ∂y σx y =

(22.4)

∂2ω ∂ω ∂2φ +y − (1 − 2ν) ; σzx = σzy = 0 , ∂x∂ y ∂x∂ y ∂x

which describe a state of plane strain, as defined in Chapter 3. This formulation is useful for plane strain problems with displacement boundary conditions8 .

22.6 A Three-dimensional Complex-Variable Solution We saw in Part IV that the complex-variable notation provides an elegant and powerful way to represent two-dimensional stress fields, and it is natural to ask whether any similar techniques could be extended to three-dimensional problems. Green9 used a solution of this kind in the development of higher order solutions for elastic plates and we shall see in Chapter 30 that similar techniques can be used for the prismatic bar with general loading on the lateral surfaces. We combine the displacement and stress components in the same way as in Chapter 20 by defining u = u x + ıu y ; Θ = σx x + σ yy Φ = σx x + 2ıσx y − σ yy ; Ψ = σzx + ıσzy .

(22.5)

Following the convention of §20.1 and §22.2, we write the components of the Papkovich-Neuber vector potential as ψ = ψx + ıψ y ; ω = ψz and hence

¯ + zω , r · ψ = ζψ + ζψ

from (20.3). Using these results and (20.2) in (21.12), we obtain the in-plane displacements as 2μu = 2 8

∂φ ∂ψ ∂ψ ∂ω − (3 − 4ν)ψ + ζ + ζ¯ + 2z , ∂ ζ¯ ∂ ζ¯ ∂ ζ¯ ∂ ζ¯

(22.6)

see for example, Problems 6.3–6.5. A. E. Green (1949), The elastic equilibrium of isotropic plates and cylinders, Proceedings of the Royal Society of London, Vol. A195, pp. 533–552.

9

376

22 The Boussinesq Potentials

and the antiplane displacement as   ∂ψ ∂φ 1 ¯ ∂ψ ∂ω + +ζ − (3 − 4ν)ω . 2μu z = ζ +z ∂z 2 ∂z ∂z ∂z Once the displacements are defined, the stress components can be found from the constitutive law as in §§22.1, 22.2. Combining these as in (22.5), we obtain     2 ∂2φ ∂ψ ψ ∂ψ ∂ 1 ∂2ψ Θ = − 2 −2 + − ζ 2 + ζ¯ 2 ∂z ∂ζ 2 ∂z ∂z ∂ ζ¯ ∂2ω ∂ω − 4ν ∂z 2 ∂z 2 ∂ φ ∂ψ ∂2ψ ∂2ω ∂2ψ Φ = 4 2 − 4(1 − 2ν) + 2ζ 2 + 2ζ¯ 2 + 4z 2 ∂ ζ¯ ∂ ζ¯ ∂ ζ¯ ∂ ζ¯ ∂ ζ¯     ∂ψ ∂ψ ∂2φ ∂2ψ 1 ¯ ∂2ψ + σzz = − 2ν + ζ + ζ ∂z 2 ∂ζ 2 ∂z 2 ∂z 2 ∂ ζ¯ −z

∂2ω ∂ω − 2(1 − ν) ∂z 2 ∂z 2 ∂2ψ ∂ φ ∂ψ ∂2ψ +ζ Ψ =2 − (1 − 2ν) + ζ¯ ¯ ¯ ¯ ∂z ∂ ζ∂z ∂ ζ∂z ∂ ζ∂z 2 ∂ ω ∂ω +2z − 2(1 − 2ν) , ¯ ∂z∂ ζ ∂ ζ¯ +z

(22.7) (22.8)

(22.9)

(22.10)

¯ z), ω(ζ, ζ, ¯ z) are real three-dimensional harmonic functions and where φ(ζ, ζ, ¯ ψ(ζ, ζ, z) is a complex three-dimensional harmonic function. Notice that we can write the three-dimensional Laplace equation in the form ∂2φ ∂2φ + 4 =0; ∂z 2 ∂ζ∂ ζ¯

∂2ψ ∂2ω + 4 =0; ∂z 2 ∂ζ∂ ζ¯

∂2ψ ∂2ψ + 4 =0, ∂z 2 ∂ζ∂ ζ¯

(22.11)

using (19.9). Thus, if the functions φ, ψ depend on z, they will not generally be ¯ We shall explore categories of complex-variable holomorphic with regard to ζ or ζ. solution of equation (22.11) in §25.8 below.

Problems 22.1 Use solution E to solve the elementary torsion problem for a cylindrical bar of radius b transmitting a torque T , the surface r = b being traction free. You will

Problems

377

need to find a suitable harmonic potential function — it will be an axisymmetric polynomial. 22.2 Find a way of taking a combination of solutions A,B,E so as to satisfy identically the global conditions σzy = σzz = 0 ; all x, y, z = 0 Thus, σzx should be the only non-zero traction component on the surface. 22.3 An important class of frictional contact problems involves the steady sliding of an indenter in the x-direction across the surface of the half-space z > 0. Assuming Coulomb’s friction law to apply with coefficient of friction f , the appropriate boundary conditions are σx z = − f σzz ; σzy = 0 , (22.12) in the contact area, the rest of the surface z = 0 being traction free. Show that (22.12) are actually global conditions for this problem and find a way of combining Solutions A,B,E so as to satisfy them identically. 22.4 A large number of thin fibres are bonded to the surface z = 0 of the elastic half-space z > 0, with orientation in the x-direction. The fibres are sufficiently rigid to prevent axial displacement u x , but sufficiently thin to offer no restraint to lateral motion u y , u z . This composite block is now indented by a frictionless rigid punch. Find a linear combination of Solutions A, B and E that satisfies the global conditions of the problem and in particular find expressions for the surface values of the normal displacement u z and the normal traction σzz in terms of the one remaining harmonic potential function. 22.5 Show that if φ, ω are taken to be independent of x, a linear combination of Solutions A and B defines a state of plane strain in the yz-plane (i.e. u x = 0, σx y = σx z = 0 and all the remaining stress and displacement components are independent of x). An infinite layer 0 < z < h rests on a frictionless rigid foundation at z = 0, whilst the surface z = h is loaded by the normal traction σzz = − p0 + p1 cos(my) , where p1 < p0 . Find the distribution of contact pressure at z = 0. 22.6 A massive asteroid passes close to the surface of the Earth causing gravitational forces that can be described by a known harmonic body force potential V (x, y, z). Assuming that the affected region of the Earth can be approximated as the tractionfree half-space z > 0, find a representation of the resulting stress and displacement fields in terms of a single harmonic potential function. An appropriate strategy is:-

378

22 The Boussinesq Potentials

(i) Using the particular solution of §21.5.1, show that the potential φ must be biharmonic. (ii) Show that this condition can be satisfied by writing φ = zχ, where χ is a harmonic function, and find the relation between χ and V . (iii) Find the tractions on the surface of the half-space in terms of derivatives of χ. (iv) Superpose appropriate combinations of Solutions A and/or B so as to render this surface traction free for all functions χ. Find the resulting stress field for the special case of a spherical asteroid of radius a, for which 4πγρρ0 a 3 , V =−  3 x 2 + y 2 + (z + h)2 where γ is the universal gravitational constant, ρ, ρ0 are the densities of the asteroid and the Earth respectively, and the centre of the sphere is at the point (0, 0, −h), a height h > a above the Earth’s surface. If the Earth and the asteroid have the same density (ρ0 = ρ) and the self-weight of the Earth is assumed to cause hydrostatic compression of magnitude ρgz, where g is the acceleration due to gravity, is it possible for the passage of the asteroid to cause tensile stresses anywhere in the Earth’s crust. If so, where? 22.7 An elastic half-space is defined by the inequality z > h(x, y), where h(x, y) defines a perturbation from the plane z = 0 that is small in the sense that the magnitude of the surface slope |∇h|  1. The surface z = h is traction free, but the extremities of the half-space are subjected to a uniform biaxial tensile stress σx x = σ yy = σ0 . Show that Solution G can be used to define an approximation to the perturbation in the stress field due to the surface profile and define the boundary conditions for the harmonic potential function χ in terms of h. Solve the resulting boundary-value problem for the case where the profile is defined by the two-dimensional depression h(x, y) =

h0a2 , x 2 + a2

and hence find the location and magnitude of the maximum tensile stress. 22.8 A general two-dimensional biharmonic function can be written in the form φ=ψ+y

∂χ , ∂y

where ψ, χ are harmonic functions of x, y. Use this form to describe the Airy stress function of Chapter 4, and develop expressions for the stress components σx x , σ yy , σx y as functions of ψ, χ. Show that the resulting solution can be written in terms of a linear combination of the harmonic functions φ, ω of equations (22.4). Suggest ways of using this relationship to determine the displacements associated with a given Airy stress function φ and illustrate the procedure with a simple example.

Chapter 23

Thermoelastic Displacement Potentials

As in the two-dimensional case (Chapter 14), three-dimensional problems of thermoelasticity are conveniently treated by finding a particular solution — i.e. a solution which satisfies the field equations without regard to boundary conditions — and completing the general solution by superposition of an appropriate representation for the general isothermal problem, such as the Papkovich-Neuber solution. In this section, we shall show that a particular solution can always be obtained in the form of a strain potential — i.e. by writing 2μu = ∇φ .

(23.1)

The thermoelastic stress-strain relations (14.3, 14.4) can be solved to give σx x = λe + 2μex x − (3λ + 2μ)αT σx y = 2μex y ,

(23.2) (23.3)

etc. Using the strain-displacement relations and substituting for u from (23.1), we then find 2μe = ∇ 2 φ λ 2 ∂2φ ∇ φ+ σx x = − (3λ + 2μ)αT 2μ ∂x 2 ∂2φ , σx y = ∂x∂ y

(23.4) (23.5) (23.6)

etc., and hence the equilibrium equation (2.1) requires that Supplementary Information The online version contains supplementary material available at https://doi.org/10.1007/978-3-031-15214-6_23.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 J. R. Barber, Elasticity, Solid Mechanics and Its Applications 172, https://doi.org/10.1007/978-3-031-15214-6_23

379

380

23 Thermoelastic Displacement Potentials

∂ ∂x



λ 2 ∇ φ + ∇ 2 φ − (3λ + 2μ)αT 2μ

 =0.

Two similar equations are obtained from (2.2, 2.3) and hence the most general solution of this form must satisfy the condition ∇2φ =

2μ(3λ + 2μ)αT +C , (λ + 2μ)

where C is an arbitrary constant1 . Since we are only looking for a particular solution, we choose to set C = 0, giving ∇2φ =

2μ(1 + ν)αT 2μ(3λ + 2μ)αT = . (λ + 2μ) (1 − ν)

(23.7)

Equation (23.7) defines the potential φ due to a source distribution proportional to the given temperature, T . Such a function can be found for all T in any geometric domain2 and hence a particular solution of the thermoelastic problem in the form (23.1) can always be found. Substituting for T from equation (23.7) back into the stress equation (23.5), we obtain   (λ + 2μ) λ ∂2φ ∂2φ − ∇2φ + σx x = = − ∇2φ 2μ 2μ ∂x 2 ∂x 2 ∂2φ ∂2φ =− 2 − 2 . ∂y ∂z The particular solution can therefore be summarized in the equations 2μ(1 + ν)αT (1 − ν) 2μu = ∇φ

∇2φ =

σx x = −

(23.8) (23.9)

∂2φ ∂2φ ∂2φ ∂2φ ∂2φ ∂2φ − 2 ; σ yy = − 2 − ; σzz = − 2 − 2 2 ∂y ∂z ∂z ∂x ∂x ∂ y2 ∂2φ ∂2φ ∂2φ ; σ yz = ; σzx = . (23.10) σx y = ∂x∂ y ∂ y∂z ∂z∂x

The equivalent expressions in cylindrical polar coördinates are 2μu r = 1

∂φ 1 ∂φ ∂φ ; 2μu θ = ; 2μu z = ∂r r ∂θ ∂z

We encountered this constant already in equation (21.3) and noted that it corresponds to a state of uniform stress. 2 For example, as a convolution integral on the point source solution 1/R.

23.1 The Method of Strain Suppression

σrr =

381

1 ∂2φ ∂2φ 1 ∂φ ∂2φ 2 2 + − ∇ φ ; σ = − ∇ φ ; σ = − ∇2φ θθ zz ∂r 2 r ∂r r 2 ∂θ2 ∂z 2 σr θ

∂ = ∂r



1 ∂φ r ∂θ

 ; σθz =

1 ∂2φ ∂2φ ; σzr = , r ∂θ∂z ∂z∂r

(23.11)

where ∇ 2 φ is given by equation (22.1). In spherical polar coördinates, we have 2μu R =

∂φ 1 ∂φ 1 ∂φ ; 2μu θ = ; 2μu β = ∂R R sin β ∂θ R ∂β

σR R =

cot β ∂φ ∂2φ 1 ∂φ + − ∇2φ − ∇ 2 φ ; σθθ = ∂ R2 R ∂R R 2 ∂β

σββ =

1 ∂2φ ∂2φ 1 ∂φ 1 + 2 2 − ∇ 2 φ ; σ Rθ = R ∂R R ∂β R sin β ∂ R∂θ

σθβ =

R2

∂2φ 1 ∂φ 1 1 ∂2φ ; σβ R = − 2 , sin β ∂θ∂β R ∂β∂ R R ∂β

(23.12)

where ∇ 2 φ is given by equation (22.2).

23.1 The Method of Strain Suppression An alternative approach for obtaining a particular solution for the thermoelastic stress field is to reduce it to a body force problem using the method of strain suppression. Suppose we were to constrain every particle of the body so as to prevent it from moving — in other words, we constrain the displacement to be everywhere zero. The strains will then be identically zero everywhere and the constitutive equations (23.2, 23.3) reduce to 2μ(1+ ν)αT ; σx y = σ yz = σzx = 0 . (1 − 2ν) (23.13) Of course, these stresses will not generally satisfy the equilibrium equations, but we can always find a body force distribution that will restore equilibrium — all we need to do is to substitute the non-equilibrated stresses (23.13) into the equilibrium equation with body forces (2.4) and use the latter as a definition of p, obtaining σx x = σ yy = σzz = −(3λ + 2μ)αT = −

p=

2μ(1 + ν) α∇T . (1 − 2ν)

(23.14)

382

23 Thermoelastic Displacement Potentials

It follows that (23.13) will define the stresses in the body for the specified temperature distribution if (i) body forces (23.14) are applied and (ii) purely normal tractions are applied at the boundaries given by t=−

2μ(1 + ν) αT n , (1 − 2ν)

(23.15)

where T is the local temperature at the boundary and n is the local outward normal unit vector. To complete the solution, we must now remove the unwanted body forces by superposing the solution of a problem with no thermal strains, but with equal and opposite body forces and with surface tractions that combined with (23.15) satisfy the traction boundary conditions of the original problem. The resulting body force problem might be solved using the body force potential of §21.5.1, since (23.14) defines a conservative vector field. However, the boundaryvalue problem would then be found to be identical with that obtained by substituting the original temperature distribution into (23.7), so this does not really constitute a new method. An alternative strategy might be to represent the body force as a distribution of point forces as suggested in §13.1.1, but using the three-dimensional Kelvin solution which we shall develop in the next chapter.

23.2 Boundary-value Problems If the function φ satisfies equation (23.8), the stresses and displacements defined by equations (23.9–23.12) will satisfy the governing equation for all x, y, z, but they will not generally satisfy the displacement and/or traction boundary conditions. To solve a boundary-value problem, we therefore must superpose the general homogeneous solution — i.e. a general state of deformation that can exist in a body without thermal strains.

23.2.1 Spherically-symmetric Stresses A particularly simple solution can be obtained if the problem is spherically symmetric — i.e. if the temperature depends only on the radius R in spherical polar coördinates (R, θ, β). In this case, equation (23.8) reduces to the ordinary differential equation 1 d d 2φ 2 dφ = 2 + ∇ φ= 2 dR R dR R dR 2

which can be integrated to yield



dφ R dR 2

 =

2μ(1 + ν)αT , (1 − ν)

23.2 Boundary-value Problems

2μ(1 + ν)α dφ = dR (1 − ν)R 2

383

 T R2d R +

A , R2

(23.16)

where A is an arbitrary constant. The only non-zero stress and displacement components are 2μu R =

dφ 2 dφ d 2φ 1 dφ ; σR R = − ; σθθ = σββ = − 2 − . dR R dR dR R dR

(23.17)

In the most general spherically-symmetric problem, we must satisfy one boundary condition (e.g. normal traction or radial displacement) at each of two boundaries for a hollow sphere, but equation (23.16) contains only one arbitrary constant. An additional degree of freedom can be introduced by superposing an arbitrary state of uniform hydrostatic stress defined by σ R R = σθθ = σββ = σ0 ; 2μu R =

σ0 (1 − 2ν)R , (1 + ν)

(23.18)

all other stress and displacement components being zero. Substituting (23.16) into (23.17) and adding (23.18), we obtain

σθθ

 2μ(1 + ν)α A σ0 (1 − 2ν)R T R2d R + 2 + 2μu R = (1 − ν)R 2 R (1 + ν)  4μ(1 + ν)α 2 A σR R = − T R 2 d R − 3 + σ0 (1 − ν)R 3 R    2μ(1 + ν)α 1 A 2 = σββ = T R d R − T + 3 + σ0 . 3 (1 − ν) R R

For a hollow sphere b < R < a, the lower limit in these integrals can be taken as the inner radius b and the boundary conditions at a, b will provide two equations for the unknowns A, σ0 . For a solid sphere, the constant A must be zero to avoid a singularity at the origin.

23.2.2 More general geometries For more general three-dimensional problems, the stresses and displacements defined by equations (23.9–23.12) can be superposed on those defined by the Galerkin vector in §21.2, or on the Boussinesq potentials of Chapter 22. In this connection, we notice that if we set T = 0 in (23.8) so that φ is harmonic, then the expressions for all the stress and displacement components reduce to those for Solution A of Tables 22.1, 22.2. In other words, Solution A can be construed as the homogeneous solution of (23.8). We therefore have two ways of proceeding, analogous to the two approaches to conservative body-force problems described in §7.3. We can either

384

23 Thermoelastic Displacement Potentials

1. Find any particular solution φ P of (23.8), find the corresponding stresses and displacements from equations (23.9–23.12) and then use the boundary values to define a corrective solution that can be formulated using Solutions A, B and E. 2. Start with a superposition of equations (23.8–23.12) and Solutions B, E, and satisfy the governing equation (23.8) and the boundary conditions at the same time. In this formulation, notice that φ is now interpreted as the general solution of (23.8), including an arbitrary harmonic function whose degrees of freedom are needed in satisfying the boundary conditions. A similar approach can also be used for conservative body-force problems, using the solution of equations (21.26–21.28) in place of Solution A.

23.3 Plane Problems If φ and T are independent of z, the non-zero stress components in (23.10) reduce to σx x = −

∂2φ ∂2φ ∂2φ ; σ ; σ = = − x y yy ∂ y2 ∂x∂ y ∂x 2

(23.19)

∂2φ ∂2φ − ∂x 2 ∂ y2

(23.20)

σzz = −

and the solution corresponds to a state of plane strain. In fact, the relations (23.19) are identical to the Airy function definitions, apart from a difference of sign, and similarly, (23.8) reduces to a particular integral of (14.7). However, an important difference is that the present solution also gives explicit relations for the displacements and can therefore be used for problems with displacement boundary conditions, including those arising for multiply-connected bodies.

23.3.1 Axisymmetric problems for the cylinder We consider the cylinder b 0 is defined by T (x, y, z) = f (z) cos(mx) , where f (z) is a known function of z only. The surface z = 0 of the half-space is traction free. Show that the normal surface displacement u z (x, y, 0) can be written u z (x, y, 0) = u 0 cos(mx) , 

where

∞

u 0 = 2α(1 + ν)

e−ms f (s)ds .

0

Important Note: This problem cannot be solved using Solutions T or P, since the temperature field will only be harmonic for certain special functions f (z). To solve the problem, assume a strain potential φ of the form φ = g(z) cos(mx) , substitute into (23.8) and solve the resulting ordinary differential equation for g(z). You can then superpose appropriate potentials from solutions A and B of Table 22.1 to satisfy the traction-free boundary condition. 23.2. Due to the decay of radioactive elements, heat is generated in a spherical planet of radius a at a uniform rate of Q 0 per unit volume per unit time. Find the resulting thermal stresses in the planet if the surface R = a is traction free and the temperature field has reached a steady state. You will need the heat conduction equation (14.11). 23.3. If a ductile material is loaded in hydrostatic tension σi j = σ0 δi j , it cannot yield and is therefore expected to fail in a brittle manner at sufficiently high σ0 . Experimentally it is difficult to load a specimen in triaxial tension, but one approach is to heat a solid sphere rapidly at its outer surface, which leads to the generation of a state of triaxial tension at the centre. Show that the instantaneous value of this stress is given by   4μ(1 + ν)α a 2  R T (R) − T (0) d R , (1 − ν)a 3 0 where a is the radius of the sphere, T (R) is the instantaneous temperature at radius R, and the surface R = a is traction free. Comment on possible limitations of this approach. 23.4. During a test on an automotive disk brake, it was found that the temperature T of the disk could be approximated by the expression T =

r2 r4 5 (1 − e−τ ) − 2 (1 − e−τ ) + 4 τ e−τ , 4 2a 8a

Problems

395

where τ = t/t0 is a dimensionless time, a is the radius of the disk, r is the distance from the axis, t is time and t0 is a constant. The disk is continuous at r = 0 and all its surfaces can be assumed to be traction free. Derive expressions for the radial and circumferential thermal stresses as functions of radius and time, and hence show that the maximum value of the stress difference (σθθ −σrr ) at any given time always occurs at the outer edge of the disk. 23.5. A heat exchanger tube consists of a long hollow cylinder of inner radius a and outer radius b, maintained at temperatures Ta , Tb respectively. Find the steady-state stress distribution in the tube if the curved surfaces are traction free and plane strain conditions can be assumed. 23.6. The steady-state temperature T (ξ, η) in the quarter plane ξ ≥ 0, η ≥ 0 is determined by the boundary conditions T (ξ, 0) = 0 ; ξ > 0 T (0, η) = T0 ; 0 < η < a =0 ; η>a. Use the function ω(ζ) = ζ 1/2 to map the problem into the half-plane y > 0 and solve for T using the method of §19.6.3. Then determine the displacement throughout the quarter plane, using equation (23.25). 23.7. Use dimensional arguments and continuity of heat flux to show that the twodimensional steady-state temperature field in the half-plane y > 0 due to a line heat source Q per unit length at the origin is T (x, y) =

Q ln(r ) + C , K

where r = x 2 + y 2 is the distance from the origin, K is the thermal conductivity and C is a constant. Use this result and equation (23.25) to determine the displacement throughout the half-plane. Assume that C = 0 and neglect rigid-body displacements. In particular, determine the displacements at the surface y = 0 and verify that they satisfy Dundurs’ theorem (14.15). 23.8. A uniform heat flux qx = q0 in a large body in a state of plane strain is perturbed by the presence of a rigid non-conducting circular inclusion of radius a. The inclusion is bonded to the matrix at r = a. Find the complete stress field in the matrix and identify the location and magnitude of the maximum tensile stress. Assume plane strain conditions.

396

23 Thermoelastic Displacement Potentials

23.9. A uniform heat flux qx = q0 in a large body in a state of plane strain with elastic and thermal properties μ1 , ν1 , α1 , K 1 is perturbed by the presence of a circular inclusion of radius a with properties μ2 , ν2 , α2 , K 2 . The inclusion is bonded to the matrix at r = a. Find the complete stress field in the matrix and the inclusion and identify the location and magnitude of the maximum tensile stress in each. Assume plane strain conditions. Note: You will find the algebra for this problem formidable if you do not use Maple or Mathematica. 23.10. A rigid non-conducting body slides at velocity V in the positive x-direction over the surface of the thermoelastic half-space z > 0, causing frictional tractions σzx and frictional heating qz given by σzx = − f σzz ; σzy = 0 ; qz = V σzx , where f is the coefficient of friction. Find a solution in terms of one potential function that satisfies these global boundary conditions in the steady state. If the half-space remains in contact with the sliding body throughout the plane z = 0, we have the additional global boundary condition u z = 0. Show that a non-trivial steady-state solution to this problem can be obtained using the harmonic function Ce−mz cos(my) , where C is an arbitrary constant and m is a parameter which must be chosen to satisfy the displacement boundary condition. Comment on the physical significance of this solution. 23.11. The half-space z > 0 is bonded to a rigid body at the surface z = 0. Show that in the steady state, the surface temperature T (x, y, 0) and the normal stress σzz (x, y, 0) at the bond are related by the equation σzz (x, y, 0) = −

2μα(1 + ν)T (x, y, 0) . (3 − 4ν)

23.12. The traction-free surface of the half-space z > 0 is uniformly heated in the annular region a 0. It is easily verified that the function  ϕ=

0 −∞



dζ x2

+

y2

+ (z − ζ)2

= ln(R + z)

satisfies these requirements. Notice that this function is singular on the negative zaxis, where R = −z, but not on the positive z-axis, where R = z. In fact, it can be regarded as the potential due to a uniform distribution of sources on the negative z-axis (r = 0, z < 0). 4

The normal traction is zero everywhere except at the origin, where there is a delta function loading, i.e. σzz = −Fδ(x)δ(y).

24.2 Dimensional Considerations

403

Substituting into Solution F, we find σzz = z

∂3ϕ ∂2ϕ 3z 3 − = ∂z 3 ∂z 2 R5

and hence the surface z = 0 is free of traction except at the origin as required5 . The force applied at the origin is  ∞ F = −2π r σzz (r, h)dr 0  ∞ r dr = −6πh 3 = −2π 2 + h 2 )5/2 (r 0 and hence the stress field due to a force F in the z-direction applied at the origin is obtained from the potential ϕ=−

F ln(R + z) . 2π

(24.4)

The displacements at the surface z = 0 are of particular interest, in view of the application to contact problems6 . We have F(1 − 2ν) ∂ϕ =− ∂r 2πr F(1 − ν) ∂ϕ = . 2μu z (r, 0) = −2(1 − ν) ∂z πr 2μu r (r, 0) = (1 − 2ν)

(24.5) (24.6)

Thus, both displacements vary inversely with distance from the point of application of the force F, as indeed we could have deduced from dimensional considerations. The singularity in displacement at the origin is not serious, since in any practical application, the force will be distributed over a finite area. If the solution for such a distributed force is found by superposition using the above result, the singularity will be integrated out. We note that the displacements are bounded at infinity, in contrast to the twodimensional case (see Chapter 12). It is therefore possible to regard the point at infinity as a reference for rigid-body displacements, if appropriate. The radial surface displacement u r is negative, indicating that a force F directed into the half-space — i.e. a compressive force — causes the surrounding surface to move towards the origin7 . We can define a surface dilatation es on the surface z = 0, such that The condition σzr = 0 on the surface is guaranteed by the use of Solution F. cf. Chapter 12. 7 Most engineers would say that this is what they would intuitively expect, but the validity of this intuition is suspect, since the corresponding result for a tensile force — for which the radial displacement is directed away from the origin — is counter-intuitive. Our expectation here is probably determined by the thought that forces of either direction will stretch the surface and hence 5 6

404

24 Singular Solutions

es ≡

∂u y ∂u r ur 1 ∂u θ ∂u x + = + + . ∂x ∂y ∂r r r ∂θ

Substituting for the displacement from equation (24.5), we obtain   1 F(1 − 2ν) 1 es = − − 2 + 2 =0. 4πμ r r In other words, the normal force F does not cause any dilatation of the surface z = 0, except at the origin. However, if we draw a circle of radius a, centre the origin on the surface, it is clear that the radius of this circle gets smaller by the amount of the inward radial displacement and hence the area of the circle A increases by ΔA = 2πau r = −

F(1 − 2ν) . 2μ

(24.7)

This result is independent of a — i.e. all circles reduce in area by the same amount8 and all of this reduction must therefore be concentrated at the origin, where in a sense a small amount of the surface is lost. We can generalize this result by superposition to state that, if there is an arbitrary distribution of purely normal (compressive) traction on the surface z = 0 of the half-space, the area A enclosed by any closed curve will change by ΔA of equation (24.7), where F is now to be interpreted as the resultant of the tractions acting within the area A.

24.3 Other Singular Solutions We have already shown how the singular solution ln(R +z) can be obtained from 1/R by partial integration, which of course is a form of superposition. A whole sequence of axially symmetric solutions can be obtained in the same way. Defining φ0 =

1 , R

(24.8)

we obtain φ−1 = ln(R + z) ; φ−2 = z ln(R + z) − R 1 2 (2z − r 2 ) ln(R + z) − 3Rz + r 2 (24.9) φ−3 = 4 1  φ−4 = 3(2z 3 − 3zr 2 ) ln(R + z) + 9zr 2 − 11z 2 R + 4r 2 R ; . . . 36 draw points towards the origin, but this is a second order (non-linear) effect and cannot be admitted in the linear theory. 8 as could be argued from the fact that the surface dilatation is zero.

24.3 Other Singular Solutions

405

by the operation9

 φn−1 (r, z) =

0 −∞

φn (r, (z − ζ))dζ .

(24.10)

All of these functions are harmonic except at the origin and on the negative z-axis, where R + z → 0. A corresponding set of harmonic potentials singular only on the positive z-axis can be obtained by setting z → −z in (24.9), which is equivalent to reflecting the potentials about the plane z = 0. It is convenient to define the signs of these functions such that φ−1 = − ln(R − z) ; φ−2 = −z ln(R − z) − R 1 (24.11) φ−3 = − (2z 2 − r 2 ) ln(R − z) + 3Rz + r 2 4 1  φ−4 = − 3(2z 3 − 3zr 2 ) ln(R − z) + 9zr 2 + 11z 2 R − 4r 2 R ; . . . , 36 since with this convention, the recurrence relation between both sets of potentials is defined by ∂φn−1 φn = . (24.12) ∂z The sequence can also be extended to functions with stronger singularities by applying (24.12) to (24.8), obtaining φ1 = −

z 3z 2 1 15z 3 9z ; φ2 = 5 − 3 ; φ3 = − 7 + 5 ; . . . 3 R R R R R

(24.13)

These functions are harmonic and bounded everywhere except at the origin. We can also generate non-axisymmetric potential functions by differentiation with respect to x or y. For example, the function 3 sin(2β) cos θ ∂ z  3x z − 3 = 5 = ∂x R R 2R 3 in spherical coördinates, is a non-axisymmetric harmonic function. However, these solutions can be obtained more systematically in terms of Legendre polynomials, as will be shown in the next chapter. The integrated solutions defined by equation (24.10) can be generalized to permit an arbitrary distribution of sources along a line or surface. For example, a fairly general axisymmetric harmonic potential can be written in the form  φ(r, z) = a

9

b

g(ζ)dt  , 2 r + (z − ζ)2

(24.14)

Some care needs to be taken with the behaviour of the integral at the upper limit to obtain bounded partial integrals by this method. A more systematic method of developing this sequence will be introduced in §25.6 below.

406

24 Singular Solutions

which defines the potential due to a distribution of sources of strength g(z) along the z-axis in the range a < z < b. Such a solution defines bounded stresses and displacements in any body which does not include this line segment. When the boundary conditions of the problem are expressed in terms of a representation such as (24.14), the problem is essentially reduced to the solution of an integral equation or a set of integral equations. This technique has a long history in the solution of axisymmetric flow problems around smooth bodies and it can be used for the related elasticity problem in which a uniform tensile stress is perturbed by an axisymmetric cavity. We shall also use an adaptation of the same method for solving crack and contact problems for the half-space in Chapters 32, 33.

24.4 Image Methods As in the two dimensional case, singular solutions for the half-space can be obtained by placing appropriate image singularities outside the body. We consider the case in which the half-space z > 0 with elastic constants μ1 , κ1 is bonded to the half-space z < 0 with constants μ2 , κ2 . Suppose that a singularity such as a concentrated force exists somewhere in the half-space z > 0 and that the same singularity in an infinite body can be defined in terms of the complex Papkovich-Neuber solution of §22.6 ¯ z), ψ (0) (ζ, ζ, ¯ z), ω (0) (ζ, ζ, ¯ z). In other words these through the potentials φ(0) (ζ, ζ, are the potentials that would apply if the entire space had the same elastic constants μ1 , κ1 . Aderogba10 shows that the stress field in the composite body can then be obtained from the potentials φ

(1)

ψ

(1)

  (Aκ21 − B) (0) = φ − Aκ1 φ z→−z + ω dz 2 z→−z  

   

 (0) (0) +(H − Aκ1 ) L +C J ψ dz ψ dzdz (0)

=ψ

(0)



(0)



 − H ψ (0) z→−z

z→−z

 (0)   ∂φ ω (1) = ω (0) − Aκ1 ω (0) z→−z + 2 A ∂z z→−z  

    −(H − Aκ1 ) J + 2 A L ψ (0) z→−z ψ (0) dz

z→−z

(24.15)

z→−z

for z > 0 and

10

K. Aderogba (1977), On eigenstresses in dissimilar media, Philosophical Magazine, Vol. 35, pp. 281–292.

24.4 Image Methods

φ

(2)

407

(Aκ21 − B)Γ = (A + 1)φ + 2

  +CΓ J ψ (0) dzdz (0)





(0)

ω dz + (1 + D − Aκ1 )Γ L

ψ (2) = −DΓ ψ (0) ω (2) = −(B + 1)φ(0) − (DΓ + B + 1)J



ψ (0) dz

(0)

ψ dz

(24.16)

for z < 0, where J {ψ} =

 ∂ψ ∂ψ 1 ∂

¯ − zJ {ψ} , + ζψ + ζψ ; L{ψ} = ∂ζ 2 ∂z ∂ ζ¯

and the composite material properties are defined as Γ = C=

μ2 ; μ1

H=

(Γ − 1) ; (Γ + 1)

A=

(Γ − 1) ; (Γ κ1 + 1)

1 [2H (κ1 + 1) − Aκ1 (κ1 + 2) − B] ; 2

B=

(Γ κ1 − κ2 ) (Γ + κ2 )

D = (H − 1)(κ1 + 1)(κ2 + 1) .

The arbitrary functions implied by the partial integrals in equations (24.15, 24.16) must be chosen so as to ensure that the resulting terms are harmonic and non-singular in the corresponding half-space.

24.4.1 The traction-free half-space The special case of the traction-free half-space is recovered by setting Γ = 0. We then have (κ2 − 1) A = B = H = −1 ; C = 2 and

  (κ2 − 1) (0) φ = φ + κ φ z→−z − ω dz 2 z→−z     



 2 (κ − 1) J +(κ − 1) L + ψ (0) dzdz ψ (0) dz 2 z→−z z→−z  (0) (0) ψ = ψ + ψ z→−z  (0)   ∂φ ω = ω (0) + κ ω (0) z→−z − 2 ∂z z→−z  

    −(κ − 1) J − 2 L ψ (0) z→−z (24.17) ψ (0) dz (0)



(0)



z→−z

408

24 Singular Solutions

Example As an example, we consider the case of the traction-free half-space with a concentrated force F acting in the z-direction at the point (0, 0, a), a distance a below the surface11 . The unperturbed field for a concentrated force at the origin is given in §24.1.2 by the potential ω=−

F F ; φ=ψ=0. =− √ 8π(1 − ν)R 8π(1 − ν) r 2 + z 2

For an equal force at the point (0, 0, a) in the infinite body, we need to make use of the results of §21.3.1 with s = ka, Ψ = kω and z → (z − a). We obtain Ψ = aω and hence, using equation (21.15), F Fa   ; φ(0) = 8π(1 − ν) r 2 + (z − a)2 8π(1 − ν) r 2 + (z − a)2 (24.18) and ψ (0) = 0. In constructing the partial integral of ω (0) for (24.17), we can ensure that the resulting term is harmonic by using the results of §24.3, but we must also ensure that the resulting potential has no singularities in the half-space z > 0. We therefore modify the first of equations (24.11) by a change of origin and write ω (0) = −



ω (0) dz =

  F ln r 2 + (z − a)2 − (z − a) , 8π(1 − ν)

which after the substitution z → −z yields

  F ln r 2 + (z + a)2 + z + a . 8π(1 − ν) This function is singular on the half line r = 0, z < −a, but this lies entirely outside the half-space z > 0 as required. Using this result and (24.18) in (24.17) and substituting κ = (3−4ν), we obtain

11

This problem was first solved by R. D. Mindlin (1936), Force at a point in the interior of a semi-infinite solid, Physics, Vol. 7, pp. 195–202. A more compendious list of solutions for various singularities in the interior of the half-space was given by R. D. Mindlin and D. H. Cheng (1950), Nuclei of strain in the semi-infinite solid, Journal of Applied Physics, Vol. 21, pp. 926–930.

Problems

409

Fa F(3 − 4ν)a   + 8π(1 − ν) r 2 + (z − a)2 8π(1 − ν) r 2 + (z + a)2  F(1 − 2ν)  2 ln − r + (z + a)2 + z + a 2π F(3 − 4ν) F   − ω=− 2 2 8π(1 − ν) r + (z − a) 8π(1 − ν) r 2 + (z + a)2 Fa(z + a) − 3/2 .  4π(1 − ν) r 2 + (z + a)2 φ=

The full stress field is then obtained by substitution into equations (22.6–22.10) and it is easily verified in Maple or Mathematica that the traction components σzr , σzθ , σzz go to zero on z = 0. In the limit a → 0, we recover the Boussinesq solution of §24.2.1 in which the force F is applied at the surface of the otherwise traction-free half-space. The potentials then reduce to F F(1 − 2ν) ln (R + z) ; ω = − , φ=− 2π 2π R agreeing with (24.4) and (22.3).

Problems 24.1. Starting from the potential function solution of Problem 22.2, use dimensional arguments to determine a suitable axisymmetric harmonic singular potential to solve the problem of a concentrated tangential force F in the x-direction, applied to the surface of the half-space at the origin. 24.2. Find the stresses and displacements12 corresponding to the use of Ψ = 1/R in Solution E. Show that the surfaces of the cone, r = z tan β are free of traction and hence use the solution to determine the stresses in a conical shaft of semi-angle β0 , transmitting a torque, T . Find the maximum shear stress at the section z = c, and compare it with the maximum shear stress in a cylindrical bar of radius a = c tan β0 transmitting the same torque. Note that (i) the maximum shear stress will not generally be either σzθ or σr θ , but must be found by appropriate coördinate transformation, and (ii) it may not necessarily occur at the outer radius of the cone, r = c tan β0 . 24.3. An otherwise uniform cylindrical bar of radius b has a small spherical hole of radius a on the axis.

12

This solution is known as the centre of rotation.

410

24 Singular Solutions

Combine the function Ψ = Az/R 3 in solution E with the elementary torsion solution for the bar without a hole (see, for example, Problem 22.1) and show that, with a suitable choice of the arbitrary constant A, the surface of the hole, R = a can be made traction free. Hence deduce the stress field near the hole for this problem. Assume that b  a, so that the perturbation due to the hole has a negligible influence on the stresses near the outer surface, r = b. 24.4. The area A on the surface of an elastic half-space is subjected to a purely normal pressure p(x, y) corresponding to the total force, F. The loaded region is now expanded in a self-similar manner by multiplying all its linear dimensions by the same ratio, λ. Each point in the new contact area is subjected to a self-similar loading such that the pressure at the point (λx, λy) is C p(x, y), where the constant C is chosen to ensure that the total force F is independent of λ. Show that the deflection at corresponding points (λx, λy, λz) will then be proportional to F/λ. 24.5. The surface of the half-space z > 0 is traction free and a concentrated heat source Q is applied at the origin. Determine the stress and displacement field in the half-space in the steady state. The traction-free condition can be satisfied by using solution P of Table 23.1. You can then use dimensional arguments to determine the dependence of the heat flux on R and hence to choose the appropriate singular potential. 24.6. The otherwise traction-free surface of the elastic half-space z > 0 is loaded by a uniform tangential force F per unit length along a closed line S directed everywhere along the local outward normal to S. Show that the average normal displacement u¯ z inside the area A enclosed by S is u¯ z = −

F(1 − 2ν) . 2μ

24.7. Use Aderogba’s formula (24.17) to determine the potentials φ, ψ, ω defining the stress and displacement fields in the traction-free half-space z > 0, subjected to a concentrated force in the x-direction acting at the point (0, 0, a). 24.8. Determine the potentials φ, ψ, ω defining the stress and displacement field in the traction-free half-space z > 0 due to a centre of dilatation of strength ΔV located at the point (0, 0, a).

Chapter 25

Spherical Harmonics

The singular solutions introduced in the last chapter are particular cases of a class of functions known as spherical harmonics. In this chapter, we shall develop these functions and some related harmonic potential functions in a more formal way. In particular, we shall identify:(i) Finite polynomial potentials expressible in the alternate forms Pn (x, y, z), Pn (r, z) cos(mθ) or R n Pn (cos β) cos(mθ), where Pn represents a polynomial of degree n. (ii) Potentials that are singular only at the origin. (iii) Potentials including the factor ln(R +z) that are singular on the negative z-axis (z < 0, r = 0). (iv) Potentials including the factor ln{(R +z)/(R −z)} that are singular everywhere on the z-axis. (v) Potentials including the factor ln(r ) and/or negative powers of r that are singular everywhere on the z-axis. All of these potentials can be obtained in axisymmetric and non-axisymmetric forms. When used in solutions A,B and E of Tables 22.1, 22.2, the bounded potentials (i) provide a complete set of functions for the solid sphere, cylinder or cone with prescribed surface tractions or displacements on the curved surfaces. These problems are three-dimensional counterparts of those considered in Chapters 5, 8 and 11. Problems for the hollow cylinder and cone can be solved by supplementing the bounded potentials with potentials (v) and (iv) respectively. Axisymmetric problems for the hollow sphere or the infinite body with a spherical hole can be solved using

Supplementary Information The online version contains supplementary material available at https://doi.org/10.1007/978-3-031-15214-6_25.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 J. R. Barber, Elasticity, Solid Mechanics and Its Applications 172, https://doi.org/10.1007/978-3-031-15214-6_25

411

412

25 Spherical Harmonics

potentials (i,ii), but these potentials do not provide a complete solution to the nonaxisymmetric problem1 . Potentials (ii) and (iii) are useful for problems of the half-space, including crack and contact problems.

25.1 Fourier Series Solution The Laplace equation in spherical polar coördinates takes the form ∇2φ ≡

1 ∂2φ ∂2φ 2 ∂φ 1 ∂ 2 φ cot β ∂φ + =0. + + + ∂ R2 R ∂R R 2 ∂β 2 R 2 ∂β R 2 sin2 β ∂θ2

(25.1)

We first perform a Fourier decomposition with respect to the variable θ, giving a series of terms involving sin(mθ), cos(mθ), m = 0, 1, 2, . . . , ∞. For the sake of brevity, we restrict attention to the cosine terms, since the sine terms will be of the same form and can be reintroduced at the end of the analysis. We therefore assume that a potential function φ can be written φ=

∞ 

f m (R, β) cos(mθ) .

(25.2)

m=0

Substituting this series into (25.1), we find that the functions f m must satisfy the equation m2 f 1 ∂2 f 2 ∂f cot β ∂ f ∂2 f − =0. (25.3) + + + 2 2 2 2 2 ∂R R ∂R R ∂β R 2 ∂β R sin β

25.2 Reduction to Legendre’s Equation We now cast equation (25.3) in terms of the new variable x = cos β , for which we need the relations  ∂ ∂x ∂ ∂ ∂ = = − sin β = − 1 − x2 ∂β ∂x ∂β ∂x ∂x    2  ∂2 ∂ ∂ ∂ 2 ∂ 2 2 − = (1 − x . = − ) −x 1 − x 1 − x ∂β 2 ∂x ∂x ∂x 2 ∂x Substituting into (25.3), we obtain 1

See §27.1.

25.3 Axisymmetric Potentials and Legendre Polynomials

m2 f 2x ∂ f (1 − x 2 ) ∂ 2 f ∂2 f 2 ∂f − − + + =0. ∂ R2 R ∂R R 2 (1 − x 2 ) R 2 ∂x R2 ∂x 2

413

(25.4)

Finally, noting that (25.4) is homogeneous in powers of R, we expand f m (R, x) as a power series — i.e. f m (R, x) =

∞ 

R n gmn (x) ,

(25.5)

n=−∞

which on substitution into equation (25.4) requires that the functions gmn (x) satisfy the ordinary differential equation (1 − x 2 )

  m2 d2g dg + n(n + 1) − g=0. − 2x dx2 dx (1 − x 2 )

(25.6)

This is the standard form of Legendre’s equation2 .

25.3 Axisymmetric Potentials and Legendre Polynomials In the special case where m = 0, the functions (25.2) will be independent of θ and hence define axisymmetric potentials. Equation (25.6) then reduces to (1 − x 2 )

d2g dg + n(n + 1)g = 0 , − 2x dx2 dx

(25.7)

one of whose solutions is a polynomial of degree n, known as a Legendre polynomial Pn (x). It can readily be determined by writing g as a general polynomial of degree n > 0, substituting into equation (25.7) and equating coefficients. The first few Legendre polynomials are P0 (x) = 1 P1 (x) = x = cos β  1 1 3 cos(2β) + 1 P2 (x) = (3x 2 − 1) = 2 4  1 1 3 P3 (x) = (5x − 3x) = 5 cos(3β) + 3 cos β (25.8) 2 8  1 1  P4 (x) = (35x 4 − 30x 2 + 3) = 35 cos(4β) + 20 cos(2β) + 9 8 64  1 1  5 3 P5 (x) = (63x − 70x + 15x) = 63 cos(5β) + 35 cos(3β) + 30 cos β , 8 128 2

See for example I. S. Gradshteyn and I. M. Ryzhik, Tables of Integrals, Series and Products, Academic Press, New York, 1980, §8.70.

414

25 Spherical Harmonics

where we have also used the relation x = cos β to express the results in terms of the polar angle β. The sequence can be extended to higher values of n by using the recurrence relation3 (n + 1)Pn+1 (x) − (2n + 1)x Pn (x) + n Pn−1 (x) = 0 (25.9) or the differential definition Pn (x) =

1 dn 2 (x − 1)n . 2n n! d x n

(25.10)

In view of the above derivations, it follows that the axisymmetric functions R n Pn (cos β)

(25.11)

are harmonic for all n. These functions are known as spherical harmonics.

25.3.1 Singular spherical harmonics Legendre polynomials are defined only for n ≥ 0, so the corresponding spherical harmonics (25.11) will be bounded at the origin, R = 0. However, if we set n = −p − 1 , where p ≥ 0, we find that n(n + 1) = (− p − 1)(− p) = p( p + 1) and hence equation (25.6) becomes d2g dg + (1 − x ) 2 − 2x dx dx 2



 m2 p( p + 1) − g=0. (1 − x 2 )

In other words, the equation is of the same form, with p replacing n. It follows that (25.12) R −n−1 Pn (cos β) are harmonic functions, which are singular at R = 0 for n ≥ 0. If n is allowed to take all non-negative values, the two expressions (25.11, 25.12) define axisymmetric harmonics for all integer powers of R, as indicated formally by the series (25.5). These functions are generated by the Maple and Mathematica files ‘sp0’, using the recurrence relation (25.9). In combination with solutions A,B and E of Chapter 22, they provide a general solution to the problem of a solid or hollow sphere loaded by arbitrary axisymmetric tractions. 3

See I. S. Gradshteyn and I. M. Ryzhik, loc. cit. §§8.81, 8.82, 8.91 for this and more results concerning Legendre functions and polynomials.

25.4 Non-axisymmetric Harmonics

415

25.3.2 Special cases If we set n = 0 in (25.12), we obtain φ = P0 (z/R)R −1 =

1 , R

which we recognize as the source solution of the previous chapter. Furthermore, if we next set n = 1 in the same expression, we obtain φ = P1 (z/R)R −2 =

z , R3

which is proportional to the function φ1 of (24.13). Similar relations exist between the sequence φ2 , φ3 , . . . , and the higher order singular spherical harmonics.

25.4 Non-axisymmetric Harmonics When m = 0, equation (25.2) will define harmonic functions that are not axisymmetric. Corresponding solutions of Legendre’s equation (25.6) can be developed from the axisymmetric forms by the relations dm Pnm (x) = (−1)m (1 − x 2 )m/2 m Pn (x) ; dx  1  1 = (1 − x 2 )−m/2 ... Pn (x)(d x)m ; x

m≤n

(25.13)

m>n.

(25.14)

x

The functions Pnm (x) are known as Legendre functions. For problems of the sphere, only the harmonics with m ≤ n are useful, since the others involve singularities on either the positive or negative z-axis. The first few non-axisymmetric Legendre functions are 1

P11 (x) = −(1 − x 2 ) 2 = − sin β 3 1 P21 (x) = −3x(1 − x 2 ) 2 = − sin(2β) 2  3 P22 (x) = 3(1 − x 2 ) = 1 − cos(2β) 2  3 3 1 1 2 P3 (x) = − (5x − 1)(1 − x 2 ) 2 = − sin β + 5 sin(3β) 2 8  15  2 2 P3 (x) = 15x(1 − x ) = cos β − cos(3β) 4  15  3 2 23 P3 (x) = −15(1 − x ) = − 3 sin β − sin(3β) . 4

416

25 Spherical Harmonics

The bounded Legendre functions (25.13) satisfy the orthogonality condition  1 (n + m)! 2 Pnm (x)Ppq (x)d x = (25.15) δnp δmq . (2n + 1) (n − m)! −1 Thus if a general harmonic function for the solid or hollow sphere is written as a double series comprising the spherical harmonics R n Pnm (cos β) cos(mθ) ; R n Pnm (cos β) sin(mθ) R Pnm (cos β) cos(mθ) ; R −n−1 Pnm (cos β) sin(mθ) −n−1

(25.16) (25.17)

m ≤ n, the coefficients in the series can be found explicitly as integrals of the boundary values4 . In combination with solutions A,B and E of Chapter 22 these functions can be used for problems of a solid or hollow sphere loaded by non-axisymmetric tractions, but the solution is complete only for the solid sphere. The functions (25.16, 25.17) are generated by the Maple and Mathematica files ‘spn’.

25.5 Cartesian and Cylindrical Polar Coördinates The bounded Legendre polynomial solutions (25.11, 25.16) also correspond to finite polynomial functions in cylindrical polar coördinates and hence can be used for the problem of the solid circular cylinder loaded by polynomial tractions on its curved boundaries. For example, the axisymmetric functions (25.11) take the form R n Pn (z/R)

(25.18)

and the first few can be expanded in cylindrical polar coördinates as R 0 P0 (z/R) = 1 R 1 P1 (z/R) = z

1

1 2 R 2 P2 (z/R) = 3z − R 2 = 2z 2 − r 2 2 2

1

1 3 3 2 5z − 3z R = 2z 3 − 3zr 2 (25.19) R P3 (z/R) = 2 2

1

1 4 R 4 P4 (z/R) = 35z − 30z 2 R 2 + 3R 4 = 8z 4 − 24z 2 r 2 + 3r 4 8 8

1

1 63z 5 − 70z 3 R 2 + 15z R 4 = 8z 5 − 40z 3r 2 + 15zr 4 , R 5 P5 (z/R) = 8 8 where we have used the relation R 2 =r 2 +z 2 . This sequence of functions is generated in the Maple and Mathematica files ‘cyl0’. 4

See Problem 25.2.

25.6 Harmonic Potentials with Logarithmic Terms

417

These functions can also be expanded as finite polynomials in Cartesian coördinates, using the relations x =r cos θ, y =r sin θ. For example, the function R 3 P31 (z/R) sin θ expands as  3 3 R 3 P31 (z/R) sin θ = − (5z 2 − R 2 ) R 2 − z 2 sin θ = − (4z 2 − r 2 )r sin θ 2 2 3 2 2 3 = − (4z y − x y − y ) . 2 Of course, these polynomial solutions can also be obtained by assuming a general polynomial form of the appropriate order and substituting into the Laplace equation to obtain constraint equations, by analogy with the procedure for biharmonic polynomials developed in §5.1.

25.6 Harmonic Potentials with Logarithmic Terms Equation (25.7), is a second order ordinary differential equation and as such must have two linearly independent solutions for each value of n, one of which is the Legendre polynomial Pn (x). To find the other solution, we define a new function h(x) through the relation (25.20) g(x) = Pn (x)h(x) and substitute into (25.7), obtaining (1 − x 2 )Pn (x)h  (x) + 2[(1 − x 2 )Pn (x) − x Pn (x)]h  (x) + [(1 − x 2 )Pn (x) − 2x Pn (x) + n(n + 1)Pn (x)]h(x) = 0 . The last term in this equation must be zero, since Pn (x) satisfies (25.7). We therefore have (1 − x 2 )Pn (x)h  (x) + 2[(1 − x 2 )Pn (x) − x Pn (x)]h  (x) = 0 .

(25.21)

This is a homogeneous first order ordinary differential equation for h  , which can be solved by separation of variables. If the non-constant solution of (25.21) is substituted into (25.20), we obtain the logarithmic Legendre function Q n (x), the first few such functions being     1+x 1+x 1 x ; Q 1 (x) = ln −1 Q 0 (x) = ln 2 1−x 2 1−x   1 1+x Q 2 (x) = (3x 2 − 1) ln 4 1−x   5x 2 2 1 1+x 3 − + Q 3 (x) = (5x − 3x) ln 4 1−x 2 3

(25.22)

418

25 Spherical Harmonics

  35x 3 1 1+x 55x 4 2 − Q 4 (x) = (35x − 30x + 3) ln + 16 1−x 8 24   4 63x 49x 2 8 1 1 + x (63x 5 − 70x 3 + 15x) ln − + − . Q 5 (x) = 16 1−x 8 8 15 The recurrence relation (n + 1)Q n+1 (x) − (2n + 1)x Q n (x) + n Q n−1 (x) = 0

(25.23)

can be used to extend this sequence to higher values of n. The corresponding harmonic functions are logarithmically singular at all points on the z-axis. For example   R+z 1 R Q 0 (z/R) = ln 2 R−z   R+z z 1 R Q 1 (z/R) = ln −R 2 R−z   3Rz 1 2 R+z 2 2 − R Q 2 (z/R) = (3z − R ) ln 4 R−z 2 0

(25.24)

√ and the factor (R −z) → 0 on the positive z-axis (r = 0, z > 0), where R = √r 2 +z 2 → z, whilst (R +z) → 0 on the negative z-axis (r = 0, z < 0), where R = r 2 +z 2 → −z. These potentials in combination with the bounded potentials (25.11), permit a general solution of the problem of a hollow cone loaded by prescribed axisymmetric polynomial tractions on the curved surfaces. They are generated in the Maple and Mathematica files ‘Qseries’, using the recurrence relation (25.23). The attentive reader will recognize a similarity between the logarithmic terms in this series and the functions φ−1 , φ−2 , . . . of equation (24.9), obtained from the source solution by successive partial integrations. However, the two sets of functions are not identical. In effect, the functions in (24.9) correspond to distributions of sources along the negative z-axis, whereas those in (25.24) involve distributions of sources along both the positive and negative z-axes. The functions (24.9) are actually more useful, since they are harmonic throughout the region z > 0 and hence can be applied to problems of the half-space with concentrated loading, as we discovered in §24.2.1. The functions Q n (x) can all be written in the form   1 1+x Q n (x) = Pn (x) ln − Wn−1 (x) , 2 1−x where Wn−1 (x) =

n  1 k=1

k

Pk−1 (x)Pn−k (x)

25.6 Harmonic Potentials with Logarithmic Terms

419

is a finite polynomial of degree of degree (n−1). In other words, the multiplier on the logarithmic term is the Legendre polynomial Pn (x) of the same order. This is also true for the functions of equation (24.9), as can be seen by comparing them with equations (25.24). A convenient way to extend the sequence of functions (24.9) is therefore to assume a solution of the form φ−n−1 = R n Pn (z/R) ln(R + z) + Rn (R, z) ,

(25.25)

where Rn is a general polynomial of degree n in R, z, substitute into the Laplace equation 1 ∂2φ ∂2φ ∂ 2 φ 1 ∂φ + + + 2 =0, (25.26) ∇2φ = ∂r 2 r ∂r r 2 ∂θ2 ∂z and use the resulting equations to determine the coefficients in Rn . This method is used in the Maple and Mathematica files ‘sing’.

25.6.1 Logarithmic functions for cylinder problems The function ln(R +z) corresponds to a uniform distribution of sources along the negative z-axis. A similar distribution along the positive z-axis would lead to the related harmonic function ln(R −z). Alternatively, we could add these two functions, obtaining the solution for a uniform distribution of sources along the entire z-axis (−∞ < z < ∞) with the result ln(R + z) + ln(R − z) = ln(R 2 − z 2 ) = 2 ln(r ) . Not surprisingly, this function is independent of z. It is actually the source solution for the two-dimensional Laplace equation 1 ∂2φ ∂ 2 φ 1 ∂φ + + =0 ∂r 2 r ∂r r 2 ∂θ2 and could have been obtained directly by seeking a singular solution of this equation that was independent of θ. More importantly, a similar superposition can be applied to the functions (25.25) to obtain a new class of logarithmically singular harmonic functions of the form φ = R n Pn (z/R) ln(r ) + Sn (r, z) ,

(25.27)

where Sn is a polynomial of degree n in r, z. These functions, in combination with the bounded harmonics of equation (25.19) and solutions A,B and E, provide a general solution to the problem of a hollow circular cylinder loaded by axisymmetric polynomial tractions on the curved surfaces. As before, the sequence (25.27) is most

420

25 Spherical Harmonics

conveniently obtained by assuming a solution of the given form with Sn a general polynomial of r, z, substituting into the Laplace equation (25.26), and using the resulting equations to determine the coefficients in Sn . This method is used in the Maple and Mathematica files ‘hol0’. The first few functions in the sequence are ϕ0 = ln(r ) ϕ1 = z ln(r ) 1 ϕ2 = [(2z 2 − r 2 ) ln(r ) + r 2 ] 2 1 ϕ3 = [(2z 3 − 3zr 2 ) ln(r ) + 3zr 2 ] 2  1 9r 4 (8z 4 − 24z 2 r 2 + 3r 4 ) ln(r ) − + 24z 2 r 2 ϕ4 = 8 2  1 45zr 4 5 3 2 4 3 2 (8z − 40z r + 15zr ) ln(r ) − + 40z r . ϕ5 = 8 2

(25.28)

Notice that only even powers of r can occur in these functions.

25.7 Non-axisymmetric Cylindrical Potentials As in §25.4, we can develop non-axisymmetric harmonic potentials in cylindrical polar coördinates in the form

φm = f m (r, z)

sin(mθ) , cos(mθ)

(25.29)

with m = 0. Substitution in (25.26) shows that the function f m must satisfy the equation m 2 fm 1 ∂ fm ∂ 2 fm ∂ 2 fm − + + =0. (25.30) ∂r 2 r ∂r r2 ∂z 2 A convenient way to obtain such functions is to note that if φm is harmonic, the function

(25.31) φm+1 = f m+1 (r, z) cos (m + 1)θ will also be harmonic if f m+1

m fm ∂ ∂ fm − ≡ rm = ∂r r ∂r



fm rm

 ,

which can also be written f m = r m Lm f 0

where

L{·} =

1 ∂ . r ∂r

(25.32)

25.7 Non-axisymmetric Cylindrical Potentials

421

To prove this result, we first note that ∂ 2 fm 1 ∂ fm ∂ 2 fm m 2 fm = − − , + ∂z 2 ∂r 2 r ∂r r2

(25.33)

from (25.30). We then substitute (25.32) into (25.31) and the resulting expression into (25.26), using (25.33) to eliminate the derivatives with respect to z. The coefficients of all the remaining derivatives will then be found to be identically zero, confirming that (25.31) is harmonic. The relation (25.32) can be used recursively to generate a sequence of non-axisymmetric harmonic potentials, starting from m = 0 with any of the axisymmetric functions developed above. For example, starting with the axisymmetric function φ0 = f 0 = R 4 P4 (z/R) =

1 4 8z − 24z 2 r 2 + 3r 4 , 8

from equation (25.19), we can construct

3 f1 ∂ f0 ∂ f1 = − 4z 2 r + r 3 ; f 2 = − = 3r 2 , f1 = ∂r 2 ∂r r from which we obtain the harmonic potentials

3 φ1 = − 4z 2 r + r 3 cos θ ; φ2 = 3r 2 cos(2θ) . 2 Similarly, from the logarithmically singular function  1 2 (2z − r 2 ) ln(r ) + r 2 φ0 = f 0 = ϕ2 = 2 of equation (25.28), we obtain f1 =

z2 r f1 2z 2 ∂ f0 ∂ f1 = −r ln(r ) + + ; f2 = − = −1 − 2 , ∂r r 2 ∂r r r

defining the potentials     r z2 2z 2 + cos θ ; φ2 = −1 − 2 cos(2θ) . φ1 = −r ln(r ) + r 2 r This process is formalized in the Maple and Mathematica files ‘cyln’ and ‘holn’, which generate a sequence of functions Fmn , Smn such that the potentials Fmn cos(mθ) ;

Fmn sin(mθ) ;

Smn cos(mθ) ;

Smn sin(mθ)

(25.34)

are harmonic. These files are easily extended to larger values of m or n if required.

422

25 Spherical Harmonics

For convenience, we list here the first few functions of this sequence for the case m = 1.  3 2 4z r − r 3 F11 = r ; F12 = 3zr ; F13 = 2   5 3 15  4 3 4z r − 3zr ; F15 = 8z r − 12z 2 r 3 + r 5 . F14 = 2 8 1 z z2 r ; S11 = ; S12 = −r ln(r ) + + r r r 2 z3 3zr S13 = −3zr ln(r ) + + r 2 4  3 3 z 15r 3 r − 4z 2 r ln(r ) + + 3z 2 r − S14 = 2 r 8 5 3  5 3 z 75zr 3zr − 4z 3r ln(r ) + + 5z 3r − . S15 = 2 r 8

(25.35)

S10 =

(25.36)

25.8 Spherical Harmonics in Complex-variable Notation In §22.6 we developed a version of the Papkovich-Neuber solution in which the ¯ z), where ζ = x +ı y, ζ¯ = x − Cartesian coördinates (x, y, z) were replaced by (ζ, ζ, ı y. In this system, the Laplace equation takes the form ∇2ψ ≡ 4

∂2ψ ∂2ψ =0. + ∂z 2 ∂ζ∂ ζ¯

(25.37)

We consider solutions of the series form ¯ z) = ψ2k (ζ, ζ,

k  ¯ z 2 j f k− j (ζ, ζ) . (2 j)! j=0

(25.38)

Substitution into (25.37) then shows that k k  4z 2 j ∂ 2 f k− j  z 2 j−2 f k− j =0 + (2 j)! ∂ζ∂ ζ¯ (2 j − 2)! j=0 j=1

and equating coefficients of z 2k−2i−2 , we have ∂ 2 f i+1 fi = − ; i = (0, k − 1) , 4 ∂ζ∂ ζ¯ ∂ 2 f0 =0, ∂ζ∂ ζ¯

(25.39) (25.40)

25.8 Spherical Harmonics in Complex-variable Notation

423

which defines a recurrence relation for the functions f i . Furthermore, equation (25.40) is the two-dimensional Laplace equation whose general solution can be written ¯ , (25.41) f 0 = g1 (ζ) + g 2 (ζ) where g1 , g 2 are general holomorphic functions of the complex variable ζ and its ¯ respectively. Thus, the most general solution of the form (25.38) can be conjugate ζ, constructed by successive integrations of (25.39), starting from (25.41). This defines an even function of z, but it is easily verified that the odd function ¯ z) = ψ2k+1 (ζ, ζ,

k  ¯ z 2 j+1 f k− j (ζ, ζ) (2 j + 1)! j=0

(25.42)

also satisfies (25.37), where the functions f i again satisfy the recurrence relations (25.39, 25.40).

25.8.1 Bounded cylindrical harmonics The bounded harmonics of equations (25.18, 25.19, 25.35) can be obtained from equations (25.38, 25.42) by choosing ¯ = g1 (ζ) = ζ m . f 0 (ζ, ζ) Successive integrations of equation (25.39) then yield ¯ = f 1 (ζ, ζ)

ζ m+1 ζ¯ ; (−4)(m + 1)

¯ = f 2 (ζ, ζ)

or in general ¯ = f i (ζ, ζ)

ζ m+2 ζ¯ , (−4)2 (m + 1)(m + 2)(2) 2

¯ i ζm m!(ζ ζ) , (−4)i (m + i)! i!

where we have omitted the arbitrary functions of integration. Substituting f i into (25.38, 25.42), we can then construct the harmonic functions χm 2k =

k  j=0

χm 2k+1

=

k  j=0

¯ k− j ζ m m!z 2 j (ζ ζ) + k − j)!(k − j)!(2 j)!

(−4)k− j (m

¯ k− j ζ m m!z 2 j+1 (ζ ζ) , (−4)k− j (m + k − j)!(k − j)!(2 j + 1)!

(25.43)

424

25 Spherical Harmonics

the first few functions in the sequence being m ¯ χm 0 (ζ, ζ, z) = ζ m ¯ z) = zζ m χ1 (ζ, ζ, ¯ m (ζ ζ)ζ z2ζ m ¯ + (25.44) χm 2 (ζ, ζ, z) = 2! (−4)(m + 1) ¯ m z3ζ m z(ζ ζ)ζ ¯ χm + 3 (ζ, ζ, z) = 3! (−4)(m + 1) 4 m ¯ 2ζ m ¯ m z 2 (ζ ζ)ζ (ζ ζ) ζ z m ¯ z) = + + . χ4 (ζ, ζ, 4! (−4)(m + 1)(2!) (−4)2 (m + 1)(m + 2)(2)

The functions (25.43, 25.44) are three-dimensionally harmonic and satisfy the recurrence relations ∂χm n = χm n−1 ; ∂z or

∂χm n = mχm−1 ; n ∂ζ



 χm n dz

=

χm n+1

;

χm n dζ

χm+1 ∂χm n−2 n . =− 4(m + 1) ∂ ζ¯

χm+1 n ; = (m + 1)



(25.45)

m−1 ¯ χm n d ζ = −4mχn+2 .

The factor ¯ = r eıθ r e−ıθ = r 2 (ζ ζ)

ζ m = r m (cos(mθ) + ı sin(mθ))

and

and hence the functions χm n can be expanded as functions of r, z with Fourier multipliers as in (25.34, 25.35). In fact they are related to the spherical harmonics (25.16) by the expressions ¯ χm n (ζ, ζ, z) =

(−2)m m! m+n m R Pm+n (z/R) exp(ımθ) , (n + 2m)! 

where R=

z 2 + ζ ζ¯ =



x 2 + y2 + z2 .

A few low order potentials that we shall need later are ζ ζ¯ zζ ζ¯ z2 z3 − ; χ03 = − , χ00 = 1 ; χ01 = z ; χ02 = 2 4 6 4 χ10 = ζ ; χ11 = zζ ;

χ12 =

ζ 2 ζ¯ zζ 2 ζ¯ z2ζ z3ζ − ; χ13 = − , 2 8 6 8

χ20 = ζ 2 ; χ21 = zζ 2 ; χ22 =

ζ 3 ζ¯ zζ 3 ζ¯ z2ζ 2 z3ζ 2 − ; χ23 = − . 2 12 6 12

(25.46)

Problems

425

25.8.2 Singular cylindrical harmonics A similar procedure can be used to develop complex versions of the singular harmonics of equations (25.34, 25.36), starting with the function ¯ = ζ −m ; m ≥ 1 f 0 (ζ, ζ) ¯ ; m=0. = ln(ζ ζ) Notice that in the complex-variable formulation of two-dimensional problems, ln(ζ) is generally excluded as being multivalued and hence non-holomorphic. However, ¯ = ln(ζ) + ln(ζ) ¯ = 2 ln(r ) ln(ζ ζ) is clearly a real single-valued harmonic function (being of the form (25.41). Singular potentials with m > 0 will eventually integrate up to include logarithmic terms when following the procedure (25.39). We see this for example in the real stress function versions (25.36) for m = 1.

Problems 25.1. Use the representation (25.10) and integration by parts to show that 

1

−1

Pn (x)Pm (x)d x =

2δmn , 2n + 1

and hence that the Legendre polynomials are orthogonal on the domain (−1, 1). 25.2. A general harmonic function φ(R, θ, β) in the spherical domain 0 < R < a can be written as a double series of bounded potentials in the form φ(R, θ, β) =

∞  ∞ 

Cnm R n Pnm (cos β) cos(mθ)

n=0 m=0

+

∞  ∞ 

Dnm R n Pnm (cos β) sin(mθ) .

n=0 m=1

Use the orthogonality conditon (25.15), to find the coefficients Cmn , Dmn in terms of the boundary value φ(a, θ, β) = f (θ, β). 25.3. Use equation (25.14) to evaluate the function P12 (x) and verify that it satisfies Legendre’s equation (25.6) with m = 2 and n = 1. Construct the appropriate harmonic potential function from equation (25.16) and verify that it is singular on the z-axis.

426

25 Spherical Harmonics

25.4. If U (x, y, z) is a harmonic function of Cartesian coordinates x, y, z, then we can construct additional harmonic functions by differentiation, as in §24.3. This is a form of linear superposition. Expand the bounded axisymmetric potential φ = R 5 P5 (z/R) of equation (25.19) as a function of x, y, z and then construct a new harmonic potential ∂2φ . ψ= ∂x 2 Show that ψ can be expressed as the sum of an axisymmetric harmonic potential and a non-axisymmetric potential of the form f (r, z) cos(2θ). 25.5. The functions (24.9) are singular only on the negative z-axis (r = 0, z < 0). Similar functions singular only on the positive z-axis are given in equations (24.11). Develop the first three of (25.28) by superposition, noting that ln(R +z)+ln(R −z) = 2 ln(r ). 25.6. Use the methodology of §5.1 to construct the most general harmonic polynomial function of degree 3 in Cartesian coördinates x, y, z. Decompose the resulting polynomial into a set of spherical harmonics. 25.7. Solve equation (25.21) for h(x) with n = 1 and hence show that Q 1 (x) satisfies Legendre’s equation for this case. 25.8. Express the recurrence relation of equations (25.29–25.32) in spherical polar coördinates R, θ, β. Check your result by deriving the first three functions starting from the source solution φ0 = 1/R and compare the results with the expressions derived in the Maple or Mathematica file ‘spn’.

Chapter 26

Cylinders and Circular Plates

The bounded harmonic potentials of equations (25.19) in combination with solutions A, B and E provide a complete solution to the problem of a solid circular cylinder loaded by axisymmetric polynomial tractions on its curved surfaces. The corresponding problem for the hollow cylinder can be solved by including also the singular potentials of equation (25.28). The method can be extended to non-axisymmetric problems using the results of §25.7. If strong boundary conditions are imposed on the curved surfaces and weak conditions on the ends, the solutions are most appropriate to problems of ‘long’ cylinders in which L  a, where L is the length of the cylinder and a is its outer radius. At the other extreme, where L  a, the same harmonic functions can be used to obtain three-dimensional solutions for in-plane loading and bending of circular plates, by imposing strong boundary conditions on the plane surfaces and weak conditions on the curved surfaces. As in Chapter 5, some indication of the order of polynomial required can be obtained from elementary Mechanics of Materials arguments.

26.1 Axisymmetric Problems for Cylinders If axisymmetric potentials are substituted into solutions A and B, we find that the circumferential displacement u θ is zero everywhere, as are the stress components σθr , σθz . There is therefore no torque transmitted across any cross section of the cylinder and no twist. The only force resultant on the cross section is a force F in the z-direction given by F=

 b a

0

2π

 σzz r dθdr = 2π

b

r σzz dr ,

(26.1)

a

where a, b are the inner and outer radii of the cylinder respectively.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 J. R. Barber, Elasticity, Solid Mechanics and Its Applications 172, https://doi.org/10.1007/978-3-031-15214-6_26

427

428

26 Cylinders and Circular Plates

By contrast, substitution of an axisymmetric potential into solution E yields a stress and displacement state where the only non-zero displacement and stress components are u θ , σθr , σθz . In this case, the axial force F = 0, but in general there will be a transmitted torque  T =

b



2π

 σθz r dθdr = 2π 2

a

0

b

r 2 σθz dr .

(26.2)

a

Notice that the additional power of r in equation (26.2) compared with (26.1) is the moment arm about the z-axis for the elemental force σθz r dr dθ. Axisymmetric problems are therefore conveniently partitioned into irrotational (torsionless) problems requiring solutions A and B only and rotational (torsional) problems requiring only solution E. There is a close parallel here with the partition of two-dimensional problems into in-plane problems (Chapters 5–14) and antiplane problems (Chapter 15). Notice that the rotational problem, like the antiplane problem, involves only one non-zero displacement (u θ ) and requires only a single harmonic potential function (in solution E). By contrast, the irrotational problem involves two non-zero displacements (u r , u z ) and requires two independent harmonic potentials, one each in solutions A and B. In the same way, the in-plane two-dimensional solution involves two displacements (e.g. u x , u y ) and requires a single biharmonic potential, which we know to be equivalent to two independent harmonic potentials.

26.1.1 The solid cylinder Problems for the solid cylinder are solved by a technique very similar to that used in Chapter 5. We shall find that we can always find finite polynomials satisfying polynomial boundary conditions on the curved boundaries in the strong (pointwise) sense, but that the boundary conditions on the ends can then only be satisfied in the weak sense. Corrective solutions for these end effects can be developed from an eigenvalue problem, as in Chapter 6.

A torsional problem As an example, we consider the solid cylinder (0 ≤r < a, 0 < z < L), built in at the end z = L and subjected to the traction distribution Sz ; σrr = σr z = 0 ; r = a L σzr = σzθ = σzz = 0 ; z = 0 .

σr θ =

(26.3) (26.4)

26.1 Axisymmetric Problems for Cylinders

429

In other words, the end z = 0 is traction free and the curved surfaces are loaded by a linearly varying torsional traction. This is clearly a torsional problem and elementary equilibrium arguments show that the torque must increase with z 2 from the free end. This suggests a leading term proportional to z 2 r in the stress component σzθ and we therefore use solution E with a 5th degree polynomial and below. We use the trial function1 Ψ =

 C4  4  C5  5 8z − 40z 3r 2 + 15zr 4 + 8z − 24z 2 r 2 + 3r 4 8 8  C2  2  C3  3 2z − 3zr 2 + 2z − r 2 + 2 2

(26.5)

from equation (25.19) and substitute into solution E (Table 22.1), obtaining the nonzero stress and displacement components 2μu θ = 5C5 (4z 3r − 3zr 3 ) + 3C4 (4z 2 r − r 3 ) + 6C3 zr + 2C2 r σr θ = −15C5 zr 2 − 3C4 r 2   r3 + 12C4 zr + 3C3r . σzθ = 15C5 2z 2 r − 2

(26.6) (26.7) (26.8)

The boundary condition (26.3) must be satisfied for all z, giving C5 = −

S ; C4 = 0 . 15a 2 L

(26.9)

The traction-free condition (26.4) can only be satisfied in the weak sense. Substituting (26.8) into (26.2), and evaluating the integral at z = 0, we obtain T =

πa 4 (3C3 − 5C5 a 2 ) 2

and hence the weak condition T = 0 gives C3 =

S 5C5 a 2 =− , 3 9L

(26.10)

using (26.9). The complete stress field is therefore   Szr 2 Sr r 2 2z 2 1 ; σzθ = . − − σr θ = 2 a L L 2a 2 a2 3 Notice that the constant C2 is not determined in this solution, since it describes merely a rigid-body displacement and does not contribute to the stress field. In order to determine it, we need to generate a weak form for the built-in condition at z = L. 1

Notice that the first two functions from equation (25.19) give no stresses or displacements when substituted into solution E and hence are omitted here.

430

26 Cylinders and Circular Plates

An appropriate condition is2  a 0

2π

u θ r 2 dθdr = 0 ; z = L .

0

Substituting for u θ from (26.6) and using (26.9, 26.10) we obtain C2 = 

and hence 2μu θ = S

2S L 2 3a 2

4z 3r 2zr 4L 2 r zr 3 − − + a2 L 3a 2 L 3L 3a 2

 .

A thermoelastic problem As a second example, we consider the traction-free solid cylinder 0 ≤r < a, −L < z < L, subjected to the steady-state thermal boundary conditions qr (a, z) = −q0 ; T (r, ±L) = 0 .

(26.11)

In other words, the curved surfaces r = a are uniformly heated, whilst the ends z = ±L are maintained at zero temperature. The problem is clearly symmetrical about z = 0 and the temperature must therefore be described by even Legendre polynomial functions. Using solution T (Table 23.1) to describe the thermoelastic field, we note that temperature is proportional to ∂χ/∂z and hence χ must be comprised of odd functions from equation (25.19) — in particular C3 (2z 3 − 3zr 2 ) + C1 z . (26.12) χ= 2 Substituting (26.12) into Table 23.1, yields   C3 2 (1 − ν) (6z − 3r 2 ) + C1 T = μα(1 + ν) 2 qr = −K

3(1 − ν)K C3r ∂T = ∂r μα(1 + ν)

(26.14)

μα(1 + ν)q0 , 3K a(1 − ν)

(26.15)

and hence C3 = − from (26.11, 26.14). 2

See §9.1.1.

(26.13)

26.1 Axisymmetric Problems for Cylinders

431

The constant C1 can be determined3 from the weak form of (26.11)2 — i.e. 

a

2π 0

    C1 a 2 2π(1 − ν) C3 3a 4 2 2 3L a − + =0 T (r, ±L)r dr = μα(1 + ν) 2 4 2

and hence C1 = −

μα(1 + ν)q0 (4L 2 − a 2 ) 3C3 (4L 2 − a 2 ) = . 4 4K a(1 − ν)

(26.16)

It follows that the complete temperature field is T (r, z) =

q0 (4(L 2 − z 2 ) + 2r 2 − a 2 ) , 4K a

from (26.13, 26.15, 26.16). A particular solution for the thermal stress field is then obtained by substituting (26.15, 26.16, 26.12) into Table 23.1. To complete the solution, we must superpose the homogeneous solution which is here given by solutions A and B, since axisymmetric temperature fields give dilatational but irrotational stress and displacement fields. To determine the appropriate order of polynomials to include in φ and ω, we compare Tables 22.1 and 23.1 and notice that whilst solution B involves the same order of differentials as solution T, the corresponding components in solution A involve one further differentiation, indicating the need for a higher order polynomial function. We therefore try the functions A2 B3 A4 (8z 4 − 24z 2 r 2 + 3r 4 ) + (2z 2 − r 2 ) ; ω = (2z 3 − 3zr 2 ) + B1 z . 8 2 2 (26.17) Substitution into Tables 22.1, 23.1 then yields the stress components φ=

 σrr = A4

 9r 2 2 − 6z − A2 − 3B3 (z 2 + 2νz 2 − νr 2 ) − 2ν B1 2

−3C3 (3z 2 − r 2 ) − 2C1 σr z = −6(2 A4 + ν B3 + C3 )zr σzz = 6A4 (2z 2 − r 2 ) + 2 A2 + 3B3 (22 − νr 2 + 2νz 2 ) + 2ν B1 + 6C3 z 2 . The strong traction-free boundary conditions

3

The constant C1 serves merely to set the base level for temperature and since no stresses are generated in a traction-free body subject to a uniform temperature rise, we could set C1 = 0 at this stage without affecting the final solution for the stresses. Notice however, that this choice would lead to different values for the constants A4 , A2 , B3 , B1 in the following analysis. Also, a uniform temperature rise does cause dilatation and hence non-zero displacements, so it is essential to solve for C1 if the displacement field is required.

432

26 Cylinders and Circular Plates

σrr (a, z) = 0 ; σr z (a, z) = 0 yield the three equations − 6A4 − 3B3 (1 + 2ν) = 9C3 2

9A4 a − A2 + 3B3 νa 2 − 2ν B1 = −3C3 a 2 + 2C1 2 2 A4 + ν B3 = −C3 ,

(26.18)

and another equation is obtained from the weak condition F = 0 on the ends z = ±L. Substituting for σzz into (26.1) and evaluating the integral, we obtain    (1 − ν)a 2 F = πa 2 3A4 (4L 2 − a 2 ) + 3B3 + 2ν L 2 + 2 A2 2  − B1 (1 − ν) + 6C3 L 2 = 0 .

(26.19)

Finally, solving (26.18, 26.19) for A4 , A2 , B3 , B1 and substituting the resulting expressions into (26.17) and Tables 22.1, 23.1 yields the complete stress field σrr =

q0 μα(a 2 − r 2 ) q0 μα(a 2 − 3r 2 ) q0 μα(2r 2 − a 2 ) ; σθθ = ; σzz = 4K a 4K a 2K a σr z = σr θ = σθz = 0 .

26.1.2 The hollow cylinder Problems for the hollow cylinder a 0, since at a such a maximum we would need ∇2

∂ϕ 0 — i.e. for which the contact traction is tensile. It follows that the maximum value of F(A) must occur when A is chosen so that there are no regions satisfying (31.8, 31.9) — i.e. when the original inequalities (31.6, 31.7) are satisfied everywhere5 . The above method has the formal advantage that it replaces the intractable inequality conditions by a variational statement, but in order to use it we need to have a way of determining the total load F(A) on the indenter for an arbitrary contact area6 A. We shall show in Chapter 38 how the reciprocal theorem can be used to simplify this problem7 .

31.3 Contact Problems Involving Adhesive Forces At very small length scales, the attractive forces between molecules (e.g. van der Waals forces) become significant and must be taken into account in the solution of contact problems. This subject has increased in importance in recent years with the emphasis on micro- and nano-scale systems. One approach is to apply the same arguments as in the Griffith theory of brittle fracture (§13.3.1) and postulate that the

5

For more details of this argument see J. R. Barber (1974), Determining the contact area in elastic contact problems, Journal of Strain Analysis, Vol. 9, pp. 230–232. 6 Even then, the variational problem is far from trivial. 7 See also R. T. Shield (1967), Load-displacement relations for elastic bodies, Zeitschrift für angewandte Mathematik und Physik, Vol. 18, pp. 682–693.

31.3 Contact Problems Involving Adhesive Forces

513

contact area will adopt the value for which the total energy (comprising elastic strain energy, potential energy of external forces and surface energy) is at a minimum. Perturbing about this minimum energy state, we then obtain G = Δγ ,

(31.10)

where G is the energy release rate introduced in §13.3.3 and Δγ is the interface energy of the two contacting materials — i.e. the work that must be done per unit area against interatomic forces at the interface to separate two bodies with atomically plane surfaces8 . It follows from equation (13.26) that the contact tractions will exhibit a tensile square-root singularity at the edge of the contact area characterized by a stress-intensity factor √ (31.11) K I = lim σzz (s) 2πs , s→0

where σzz (s) is the normal contact traction (tensile positive) at a distance s from the edge of the contact area. For the more general case of frictionless contact between dissimilar materials, equation (13.26) is modified to   K 2 κ1 + 1 κ2 + 1 + G= I 16 μ1 μ2 and for three-dimensional problems, the local conditions at the edge of the contact are always those of plane strain (κ = 3−4ν), so (31.10) gives  (31.12) K I = 2E ∗ Δγ , ∗

where E is the composite contact modulus defined such that 1 1 − ν1 1 − ν2 + . ∗ = 2μ1 2μ2 E

(31.13)

This approach to adhesive contact problems was first introduced by Johnson, Kendall and Roberts9 and has come to be known as the JKR theory. Notice that it predicts that the stress-intensity factor is constant around the edge of the contact region, regardless of its shape. Of course, in the special case where adhesive forces and hence surface energy are neglected, this stress-intensity factor will be zero and the theory reduces to the classical condition that the contact tractions are square-root bounded at the edge of the contact area. The JKR theory is approximate in that it neglects the attractive forces between the surfaces in the region where the gap is small but non-zero. More exact numerical solutions using realistic force separation laws show that the theory provides a good

This will generally be less than the sum of the surface energies γ1 + γ2 of the two contacting materials, except in the case of similar materials. 9 K. L. Johnson, K. Kendall and A. D. Roberts (1971), Surface energy and the contact of elastic solids, Proceedings of the Royal Society of London, Vol. A324, pp. 301–313. 8

514

31 Frictionless Contact

limiting solution in the range where the dimensionless parameter  Ψ ≡

R ε3



Δγ E∗

2 1/3 1,

where R is a representative radius of the contacting surfaces and ε is the equilibrium intermolecular distance10 . An alternative criterion, analogous to the small-scale yielding criterion for elasticplastic fracture (See §13.3.1), is that the width ∗

E Δγ s0 = πσ02 of the region in which the tensile traction is predicted to exceed the theoretical strength σ0 should be small compared with the linear dimensions of the contact area.

Problems 31.1. Using the result of Problem 24.4 or otherwise, show that if the punch profile u 0 (x, y) has the form u 0 = r p f (θ) where p > 0 and f (θ) is any function of θ, the resulting frictionless contact problems at different indentation forces will be self-similar11 . Show also that if l is a representative dimension of the contact area, the indentation force F and the rigid-body indentation d will vary according to F ∼ l p+1 ; d ∼ l p , and hence that the indentation has a stiffening load-displacement relation 1

F ∼ d 1+ p . Verify that the Hertzian contact relations (§32.2.5 below) agree with this result.

10

J. A. Greenwood (1997), Adhesion of elastic spheres, Proceedings of the Royal Society of London, Vol. A453, pp. 1277–1297. 11 D. A. Spence (1973), An eigenvalue problem for elastic contact with finite friction, Proceedings of the Cambridge Philosophical Society, Vol. 73, pp. 249–268, has shown that this argument also extends to problems with Coulomb friction at the interface, in which case, the zones of stick and slip also remain self-similar with monotonically increasing indentation force.

Problems

515

31.2. A frictionless rigid body is pressed into an elastic body of shear modulus μ and Poisson’s ratio ν by a normal force F, establishing a contact area A and causing the rigid body to move a distance δ. Use the results of §31.1 to define the boundaryvalue problem for the potential function ϕ corresponding to the incremental problem, in which an infinitesimal additional force increment ΔF produces an incremental rigid-body displacement Δδ. Suppose that the rigid body is a perfect electrical conductor at potential V0 and the ‘potential at infinity’ in the elastic body is maintained at zero. Define the boundaryvalue problem for the potential V in the elastic body, noting that the current density vector i is given by Ohm’s law 1 i = − ∇V , ρ where ρ is the electrical resistivity of the material. Include in your statement an expression for the total current I transmitted through the contact interface. By comparing the two potential problems or otherwise, show that the electrical contact resistance R = V0 /I is related to the incremental contact stiffness d F/dδ by the equation (1 − ν) d F 1 = . R ρμ dδ 31.3. Use equations (19.8, 19.10) to show that the function 

 x + ız f (x, z) =  arcsin a satisfies Laplace’s equation. Show that the function ∂ϕ = A + B f (x, z) ∂z can be used to solve the plane strain equivalent of the flat punch problem of §12.5.2 with suitable choices for the constants A, B. In particular, obtain expressions for the contact pressure distribution and for the normal surface displacement u z outside the contact area. Note: For complex arguments, we can write12 arcsin(ζ) =

 1 ln(ıζ + 1 − ζ 2 ) . ı

31.4. Show that for frictionless contact problems for the half-space z > 0, the normal stress components near the surface z = 0 satisfy the relation 12

See I. S. Gradshteyn and I. M. Ryzhik, Tables of Integrals, Series and Products, Academic Press, New York, 1980, §1.622.1.

516

31 Frictionless Contact

σx x + σ yy = (1 + 2ν)σzz , where σzz is the applied traction. 31.5. Using results from §24.2.1, find the curvature ∂2uz ∂2uz + ∂x 2 ∂z 2 of the surface of the half-space z > 0 due to the concentrated normal force F in Figure 24.2. Using this result or otherwise, show that if a smooth frictionless indenting body is everywhere convex, meaning ∂2u0 ∂2u0 + a. 0

These constitute a pair of dual integral equations for the function A(ξ). Sneddon used the method of Titchmarsh3 and Busbridge4 to reduce equations of this type to a single equation, but a more recent solution by Sneddon5 and formalized by Gladwell6 effects this reduction more efficiently for the classes of equation considered here.

32.2 Collins’ Method A related method, which has the advantage of yielding a single integral equation in elementary functions directly, was introduced by Green and Zerna7 and applied to a wide range of axisymmetric boundary-value problems by Collins8 .

32.2.1 Indentation by a flat punch To introduce Collins’ method, we first examine the simpler problem in which the punch is flat and hence u 0 (r ) is a constant. This was first solved by Boussinesq in the 1880s. A particularly elegant solution was developed by Love9 , using a series of complex harmonic potential functions generated from the real Legendre polynomial solutions

3

E. C. Titchmarsh, Introduction to the Theory of Fourier Integrals, Clarendon Press, Oxford, 1937. I. W. Busbridge (1938), Dual integral equations, Proceedings of the London Mathematical Society, Ser. 2 Vol. 44, pp. 115–129 5 I. N. Sneddon (1960), The elementary solution of dual integral equations, Proceedings of the Glasgow Mathematical Association, Vol. 4, pp. 108–110. 6 G. M. L. Gladwell, loc. cit., Chapters 5, 10. 7 A. E. Green and W. Zerna, loc. cit.. 8 W. D. Collins (1959), On the solution of some axisymmetric boundary-value problems by means of integral equations, II: Further problems for a circular disc and a spherical cap, Mathematika, Vol. 6, pp. 120–133. 9 A. E. H. Love (1939), Boussinesq’s problem for a rigid cone, Quarterly Journal of Mathematics, Vol. 10, pp. 161–175. 4

32.2 Collins’ Method

519

of Chapter 25 by substituting (z +ıa) for z. This is tantamount to putting the origin at the ‘imaginary’ point (0, −ıa). The real and imaginary parts of the resulting functions are separately harmonic and have discontinuities at r = a on the plane z = 0. For example, if we start with the harmonic function ln(R +z) from equation (24.9) and replace z by z +ıa, we can define the new harmonic function ∗

φ = φ1 + ıφ2 = ln(R + z + ıa) , where

∗

R =

 r 2 + (z + ıa)2 .

We also record the first derivative of φ which is 1 ∂φ = ∗ . ∂z R √ ∗ On the plane z = 0, R → r 2 −a 2 and hence φ(r, 0) = ln

  r 2 − a 2 + ıa

∂φ 1 (r, 0) = √ . 2 ∂z r − a2 Both the real and imaginary parts of these functions have discontinuities at r = a on the plane z = 0. For example φ2 (r, 0) ≡ {φ(r, 0)} =

π ; 2

= sin−1

0≤r a, r

(32.3) (32.4)

whilst 1 ∂φ2 (r, 0) = − √ ; 0≤r a.

(32.5) (32.6)

Comparing these results with the boundary conditions (32.1, 32.2), we see that the function   ∗  2μu 0 ∂ϕ =−  ln R + z + ıa (32.7) ∂z π(1 − ν) satisfies both boundary conditions identically for the case where the function u 0 (r ) is a constant — i.e. if the rigid indenter is flat and is pressed a distance u 0 into the half-space, as shown in Figure 32.1.

520

32 The Boundary-value Problem

Fig. 32.1 Indentation by a flat-ended cylindrical rigid punch.

The contact pressure distribution is then immediately obtained from equations (31.1, 32.5, 32.7) and is p(r ) = −σzz (r, 0) =

2μu 0 ; 0≤r a, r

and the total indenting force 

a

F = 2π 0

r p(r )dr =

4μu 0 (1 − ν)

 0

a

r dr 4μu 0 a = . √ (1 − ν) a2 − r 2

The remaining stress and displacement components can be obtained by substituting (32.7) into the corresponding expressions for Solution F from Table 22.3. This is Love’s solution of the flat punch problem. To gain an appreciation of its elegance, it is really necessary to compare it with the original Hankel transform solution10 which is extremely complicated.

10

I. N. Sneddon (1946), Boussinesq’s problem for a flat-ended cylinder, Proceedings of the Cambridge Philosophical Society, Vol. 42, pp. 29–39.

32.2 Collins’ Method

521

32.2.2 Integral representation Green and Zerna11 extended the method of the last section to give a general solution of the boundary-value problem of equations (32.1, 32.2), using the integral representation  a  g(t)dt g(t)dt ∂ϕ 1 a   = = , (32.9) 2 2 2 ∂z 2 −a r + (z + ıt)2 r + (z + ıt) 0 where g(t) is an even function of t. It can be shown that equation (32.9) satisfies (32.2) identically and the remaining boundary condition (32.1) will then yield an integral equation for the unknown function g(t). In effect, equation (32.9) is a superposition of solutions of the form (r 2 +(z + 2 −1/2 derived from the source solution 1/R. We could therefore describe ϕ as the ıt) ) potential due to an arbitrary distribution g(t) of point sources along the imaginary z-axis between (0,0) and (0, ıa). It is therefore a logical development of the classical method of obtaining axisymmetric potential functions by distributing singularities along the axis of symmetry12 . A closely related solution is given by Segedin13 , who develops it as a convolution integral of an arbitrary kernel function with the Boussinesq solution of §32.2.1. He uses this method to obtain the solution for a power law punch (u 0 (r ) ∼ r n ) and treats more general problems by superposition after expanding the punch profile as a power series in r . It should be noted that Segedin’s solution is restricted to indentation by a punch of continuous profile14 , in which case the contact pressure tends to zero at r = a. The idea of representing a general solution by superposition of Boussinesq-type solutions for different values of a has also been used as a direct numerical method by Maw et al.15 . Green’s method was extensively developed by Collins, who used it to treat many interesting problems including the indentation problem for an annular punch16 and a problem involving ‘radiation’ boundary conditions17 .

11

A. E. Green and W. Zerna, loc.cit., §§5.8–5.10. see §24.3 and particularly equation (24.14). 13 C. M. Segedin (1957), The relation between load and penetration for a spherical punch, Mathematika, Vol. 4, pp. 156–161. 14 see §31.2. 15 N. Maw, J. R. Barber and J. N. Fawcett (1976), The oblique impact of elastic spheres, Wear, Vol. 38, pp. 101–114. 16 W. D. Collins (1963), On the solution of some axisymmetric boundary-value problems by means of integral equations, VIII: Potential problems for a circular annulus, Proceedings of the Edinburgh Mathematical Society, Vol. 13, pp. 235–246. 17 W. D. Collins (1959), On the solution of some axisymmetric boundary-value problems by means of integral equations, II: Further problems for a circular disc and a spherical cap, Mathematika, Vol. 6, pp. 120–133. 12

522

32 The Boundary-value Problem

32.2.3 Basic forms and surface values To represent harmonic potential functions we shall use suitable combinations of the four basic forms  a  ∞ g1 (t)F(r, z, t)dt ; φ2 = g2 (t)F(r, z, t)dt φ1 = 0 a a ∞ g3 (t)F(r, z, t)dt ; φ4 =  g4 (t)F(r, z, t)dt (32.10) φ3 =  0

a

where F(r, z, t) = ln

  r 2 + (z + ıt)2 + z + ıt .

(32.11)

The square root in (32.11) is interpreted as  r 2 + (z + ıt)2 = ρeıυ/2 , where ρ=

 4

(r 2

+

z2

t 2 )2

−

+

4z 2 t 2

; υ = tan

−1



2zt r 2 + z2 − t 2



and ρ ≥ 0, 0 ≤ υ < π. Equations (32.10) can be written in two alternative forms which for φ1 are φ1 =

1 2



a

g1 (t){F(r, z, t) + F(r, z, −t)}dt

(32.12)

0

and φ1 =

1 2



a

−a

g1 (t)F(r, z, t)dt .

(32.13)

We note that (32.12) is exactly equivalent to (32.13) if and only if g1 is an even function of t. If the boundary values of φ, ∂φ/∂z etc. specified at z = 0 are even in r , it will be found that g1 , g2 are even and g3 , g4 odd functions of t and forms like (32.13) can be used. The majority of problems fall into this category, but there are important exceptions such as the conical punch (where u z is proportional to r ) and problems with Coulomb friction for which σzr is proportional to σzz in some region. In these problems, forms like (32.13) can only be used if gi (t) is extended into t < 0 by a definition with the required symmetry. Expressions for the important derivatives of the functions φi at the surface z = 0 are given in Table 32.1. In certain cases, higher derivatives are required — notably in thermoelastic problems, where heat flux is proportional to ∂ 3 φ/∂z 3 (see Solution P, Table 23.1). Higher derivatives are most easily obtained by differentiating within the plane z = 0, making use of the fact that for an axisymmetric harmonic function f (r, z),

32.2 Collins’ Method

523

∂2 f 1 ∂ ∂f r . =− ∂z 2 r ∂r ∂r

(32.14)

The reader can verify that this result permits the expressions for ∂ 2 φi /∂z 2 in Table 32.1 to be obtained from those for ∂φi /∂r . Table 32.1 Surface values of the derivatives of the functions φi (Equations (32.10)). 0 b2 /R and hence 8μb3 . (32.25) F≥ 3(1 − ν)R For smaller values of F, contact will occur only in a smaller circle of radius a < b. In this case, the material of the punch outside r = a plays no rôle in the problem and the contact pressure can be determined by replacing b by a in (32.24). Furthermore, the smooth contact condition requires that p(r ) be bounded at r = a and this is equivalent to enforcing the equality in (32.25) with b replaced by a, and then solving for a giving  3(1 − ν)F R 1/3 . a= 8μ

32.2 Collins’ Method

527

However, we shall introduce a more systematic approach to problems of this class in the next section.

32.2.5 Smooth contact problems If the indenter is smooth, the extent of the contact area must be found as part of the solution by enforcing the unilateral inequalities (31.6, 31.7). One way to do this is first to solve the problem assuming a is an independent variable and then determine it from the condition that p(r ) be non-singular at the boundary of the contact area r = a, as in the two-dimensional Hertz problem of §12.5.3. Alternatively, we recall from Chapter 31 that the appropriate contact area is that which maximizes the indentation load F, so for a circular contact area of radius a, dF =0 da

g1 (a) = 0 ,

and hence

(32.26)

from (32.21). However, more direct relationships can be established between the contact radius, the indenting force and the shape of the indenter, defined merely through its slope u 0 (r ). Starting with the right-hand side of equation (32.20), we integrate by parts, obtaining  a  a  tg1 (t)dt = g1 (a) a 2 − r 2 − t 2 − r 2 g1 (t)dt √ t2 − r2 r r and hence

g1 (a) + p(r ) = − √ a2 − r 2

 r

a

g (t)dt . √1 t2 − r2

The second term is bounded at r → a and hence the tractions will be bounded there if and only if g1 (a) = 0, as in (32.26). Using this result, we can then write a simpler expression for the contact pressure in smooth contact problems as  p(r ) = r

a

g (t)dt . √1 t2 − r2

(32.27)

Since the contact pressure depends only on the derivative g1 , we can also achieve some simplification by performing a similar integration by parts on the right-hand side of (32.19). We obtain  0

and hence

s

r u 0 (r )dr = u 0 (0)s + √ s2 − r 2

 s 0

s 2 − r 2 u 0 (r )dr

528

32 The Boundary-value Problem

g1 (s) = −



 s u (r )dr 2μ , d +s √0 π(1 − ν) s2 − r 2 0

(32.28)

where d = u 0 (0) represents the rigid-body indentation of the punch. We can determine d by setting g1 (a) = 0 in (32.28), giving  d = −a 0

and hence g1 (s) =

a

u (r )dr √0 a2 − r 2



 a  s u (r )dr u (r )dr 2μ . a −s √0 √0 π(1 − ν) a2 − r 2 s2 − r 2 0 0

(32.29)

(32.30)

Notice that this result depends only on the shape of the punch as defined by the derivative u 0 (r ). Also, differentiating (32.30), we obtain 

 s d u (r )dr 2μ . (32.31) s g1 (s) = − √0 π(1 − ν) ds s2 − r 2 0 Substituting (32.30) into (32.21), we have 

 a  a  s 4μ u (r )dr u 0 (r )dr ds . F =− a2 − s √0 √ (1 − ν) a2 − r 2 s2 − r 2 0 0 0 Changing the order of integration in the second term and performing the inner integral, we have  a  s  a  a  a u 0 (r )dr ds sdsdr s = u 0 (r ) = a 2 − r 2 u 0 (r )dr √ √ s2 − r 2 s2 − r 2 0 0 0 r 0 and hence, after some algebraic simplification,  a 2 r u 0 (r )dr 4μ . F =− √ (1 − ν) 0 a2 − r 2

(32.32)

In a typical problem, the shape of the punch u 0 (r ) is a known function and either the force F or the indentation d will be prescribed. The contact radius a can then be determined by evaluating the integrals in (32.32) or (32.29) and solving the resulting equation. The function g1 (s) and the contact pressure can then be obtained from equations (32.31) and (32.27) respectively. Example – The Hertz problem As an example, we consider the problem in which a half-space is indented by a rigid sphere of radius R, in which case the slope of the indenter is defined by

32.2 Collins’ Method

529

u 0 = −

r . R

Substituting in equation (32.32) and performing the integration, we obtain 4μ F= (1 − ν)R



r 3 (r )dr 8μa 3 . = √ 3(1 − ν)R a2 − r 2

a 0

(32.33)

Also, the rigid-body indentation is obtained from (32.29) as a d= R



a

r dr a2 = . √ R a2 − r 2

0

(32.34)

Alternatively, we can eliminate a between equations (32.33, 32.34), obtaining 8μR 1/2 d 3/2 , 3(1 − ν)

F=

which exhibits a stiffening characteristic with increasing indentation. For the contact pressure distribution, we first note that 

s

0

u (r )dr 1 =− √0 2 2 R s −r



s

√ 0

r dr s2

− r2

=−

s . R

Substituting this result into the right-hand side of (32.31), we then obtain g1 (s) =

4μs , π(1 − ν)R

after which (32.27) gives p(r ) =

4μ π(1 − ν)R

 r

a

√ tdt 4μ a 2 − r 2 . = √ π(1 − ν)R t2 − r2

(32.35)

32.2.6 Choice of form Each of the functions φi in Table 32.1 has a zero in ∂φ/∂z or ∂ 2 φ/∂z 2 either in 0 a. Thus, in the above examples it would have been possible to satisfy (32.2) by choosing ∂ϕ/∂z = φ3 instead of ϕ = φ1 . The choice is best made by examining the requirements of continuity imposed at r = a, z = 0 by the physical problem. It can be shown that, if gi (a) is bounded, the expressions in Table 32.1 will define continuous values of ∂φi /∂z, but ∂ 2 φi /∂z 2 will be discontinuous unless gi (a) = 0. Thus if, as in the present case, the function ∂ϕ/∂z represents a physical quantity like displacement or temperature which is required to

530

32 The Boundary-value Problem

be continuous, it is appropriate to choose ∂ϕ/∂z = ∂φ1 /∂z. The alternative choice of ∂ϕ/∂z = φ3 imposes too strong a continuity condition at r = a, since it precludes discontinuities in second derivatives and hence in the stress components. Of course, in problems involving contact between smooth surfaces, the normal traction σzz must also be continuous at the edge of the contact region. The most straightforward treatment is that given in §32.2.5, in which the continuity condition furnishes an extra condition to determine the radius of the contact region, which is not known a priori. However, it would also be possible to enforce the required continuity through the formulation by using ∂ϕ/∂z = φ3 in which case the contact radius is prescribed and the rigid-body indentation d of the punch must be allowed to float. This is essentially the technique used by Segedin19 .

32.3 Non-axisymmetric Problems The method described in this chapter can also be applied to problems in which non-axisymmetric boundary conditions are imposed interior to and exterior to the circle r = a — for example, if the punch has a profile u 0 (r, θ) that depends on both polar coördinates. The first step is to perform a Fourier decomposition of the profile, so that the boundary conditions can be expressed as a series of terms of the form f m (r ) cos(mθ) or f m (r ) sin(mθ), where m is an integer. These terms can then be treated separately using linear superposition. Copson20 shows that if a harmonic potential function φ(r, θ, z) satisfies the mixed boundary conditions φ(r, θ, 0) = f (r ) cos(mθ) ; 0 ≤ r < a ∂φ (r, θ, 0) = 0 ; r >a, ∂z the surface value of the derivative in 0 ≤r < a can be written  a tg(t)dt d ∂φ m−1 (r, θ, 0) = r cos(mθ) , √ ∂z dr r t2 − r2 where g(t) =

2 d πt 2m dt

 0

t

s m+1 f (s)ds . √ t 2 − s2

(32.36)

(32.37)

(32.38)

This result can clearly be used to solve non-axisymmetric problems of the form (32.1, 32.2) by setting 19

C. M. Segedin (1957), The relation between load and penetration for a spherical punch, Mathematika, Vol. 4, pp. 156–161. 20 E. T. Copson (1947), On the problem of the electrified disc, Proceedings of the Edinburgh Mathematical Society, Vol. 8, pp. 14–19.

32.3 Non-axisymmetric Problems

531

∂ϕ . ∂z

φ=

(32.39)

With this notation, equations (32.36, 32.37) reduce to the equivalent expressions for φ1 in Table 32.1 in the axisymmetric case m = 0.

Example: The tilted flat punch We suppose that a cylindrical rigid flat punch of radius a is pressed into the surface of the elastic half-space z > 0 by an offset force F, whose line of action passes through the point ( , 0, 0) causing the punch to tilt through some small angle α. We restrict attention to the case where the offset is sufficiently small to ensure that the complete face of the punch makes contact with the half-space. The boundary condition in 0 ≤r < a is then (32.40) u z (r, θ, 0) = d + αr cos θ and since the problem is linear, we can decompose the solution into an axisymmetric and a non-axisymmetric term, the former being identical to that given in §32.2.1. To complete the solution, we therefore seek a harmonic function ϕ satisfying the boundary conditions μαr cos θ ∂ϕ =− ; 0≤r a, z=0, ∂z 2

(32.41)

from (32.1, 32.2). Using (32.39), these conditions are equivalent to (32.36) with m = 1 and μαr . f (r ) = − (1 − ν) Equation (32.38) then gives 

2μα d g(t) = − π(1 − ν)t 2 dt

0

t

s 3 f (s)ds 4μα =− √ 2 2 π(1 − ν) t −s

and substituting into (32.37), we obtain d ∂2ϕ (r, θ, 0) = cos θ ∂z 2 dr

 r

a

tg(t)dt 4μαr cos θ = . √ √ 2 2 t −r π(1 − ν) a 2 − r 2

The contact pressure distribution under the punch is then obtained as p(r, θ) =

∂2ϕ 2μ = √ (d + 2αr cos θ) , ∂z 2 π(1 − ν) a 2 − r 2

532

32 The Boundary-value Problem

where we have superposed the axisymmetric contribution due to the term d in (32.40) using (32.8). This solution of the tilted punch problem was first given by Green21 . Contact will occur over the complete face of the punch if and only if p(r, θ) > 0 for all (r, θ) in 0 ≤r < a. This condition will fail at (a, −π), implying local separation, if d . (32.42) α> 2a The rigid-body motion of the punch defined by the constants d, α can be related to the force F and its offset by equilibrium considerations. We obtain F=

 a 0

F =

0

 a 0

2π

0

2π

4μad (1 − ν) 8μa 3 α p(r, θ)r 2 cos θdθdr = 3(1 − ν) p(r, θ)r dθdr =

and using (32.42), we conclude that for full-face contact, we require ≤

a . 3

In other words, the line of action of the indenting force must lie within a central circle of radius a/3.

32.3.1 The full stress field In his solution of the tilted punch problem, Green used a semi-intuitive method similar to that introduced in §32.2.1 to deduce the form of the harmonic potential ϕ throughout the half-space from the surface values obtained above. A more formal approach is to use the recurrence relation (25.31, 25.32) to relate the non-axisymmetric problem to an axisymmetric problem that can be solved using the method of §32.2.2 and Table 32.1. For example, in the case of the tilted punch problem, the non-axisymmetric term in ϕ(r, θ, z) can be written f 1 (r, z) cos θ, where f 1 satisfies the boundary conditions μαr ∂ f1 =− ; 0≤r a, z=0, ∂z 2

21

(32.43)

A. E. Green (1949), On Boussinesq’s problem and penny-shaped cracks, Mathematical Proceedings of the Cambridge Philosophical Society, Vol. 45, pp. 251–257.

32.3 Non-axisymmetric Problems

533

from (32.41). Setting m = 0 in (25.32), we can define an axisymmetric harmonic function f 0 (r, z) such that ∂ f0 f 1 (r, z) = . (32.44) ∂r Substituting this relation into (32.43)1 and integrating with respect to r , we conclude that f 0 must satisfy the boundary condition ∂ f0 (r, 0) = ∂z



∂ f1 μαr 2 (r, 0)dr = − +C ; 0 ≤r a, z=0. (32.45) ∂z 2 Equation (32.45) can be satisfied identically by representing the axisymmetric harmonic f 0 (r, z) in the form φ1 of equation (32.10). Table 32.1 then gives  0

r

g1 (t)dt μαr 2 +C =− √ 2(1 − ν) r2 − t2

and the Abel equation solution is obtained from Table 32.2 as 2 d g1 (x) = π dx

 0

x

  μαt 3 dt μαx 2 2 − + Ct √ − +C . = 2(1 − ν) π (1 − ν) x2 − t2

The constant C must be chosen so as to satisfy the condition of continuity of displacement at r = a. As explained in §32.2.6, the representations (32.10) ensure continuity of the derivative ∂φi /∂z at r = a as long as the corresponding function gi (t) is bounded. In the present case, this means that ∂ f 0 /∂z will be continuous for all values of C. However, the differentiation in (32.44) will introduce a square-root singularity in ∂ f 1 /∂z unless the constant C is chosen so as to satisfy the stronger condition g1 (a) = 0, giving 2μα(a 2 − x 2 ) . g1 (x) = π(1 − ν) The function f 0 (r, z) is then given by (32.10)1 with this value of g1 after which f 1 (r, z) is defined by (32.44). For higher-order problems [larger values of m in ϕ = f m (r, z) cos(mθ)], additional constants of integration will appear in the boundary conditions for f 0 (r, z) and stronger continuity conditions must be imposed at r = a. In fact g1 (t) and its first m −1 derivatives must be set to zero at t = a.

534

32 The Boundary-value Problem

Problems 32.1. A harmonic potential function ω(r, z) in the region z > 0 is defined by the surface values ω(r, 0) = A(a 2 −r 2 ) ; 0 ≤r < a, ω(r, 0) = 0 ; r > a. Select a suitable form for the function using Table 32.1 and then use the boundary conditions to determine the corresponding unknown function gi (t). 32.2. Use equation (32.14) to derive the expressions for ∂ 2 φ1 /∂z 2 in Table 32.1 from those for ∂φ1 /∂r . 32.3. An elastic half-space is indented by a rigid cylindrical punch of radius a with a concave spherical end of radius R a. Find the contact pressure distribution p(r ) and hence determine the minimum force F0 required to maintain contact over the entire punch surface. Do not attempt to solve the problem for F < F0 . 32.4. A rigid flat punch of radius a is bonded to the surface of the elastic half-space z > 0. A torque T is then applied to the punch. Use Solution E to describe the stress field in the half-space and Collins’ method to solve the resulting boundary-value problem for Ψ . In particular, find the distribution of traction between the punch and the half-space and the rotation of the punch due to the torque. 32.5. A rigid sphere of radius R is pressed into an elastic half-space of shear modulus μ and Poisson’s ratio ν. Assuming that the interface energy for this material combination is Δγ, find a relation between the indenting force F and the radius of the contact area a. You should use the conditions (31.11, 31.12) to determine the unknown constant d in (32.22) and then (32.21) to determine F. Hence determine the contact radius when the sphere is unloaded (F = 0) and the tensile force required to pull the sphere out of contact22 . 32.6. (i) An elastic half-space is indented by a conical frictionless rigid punch as shown in Figure 32.3, the penetration of the apex being d. Find the contact pressure distribution, the total load F and the contact radius a. The semi-angle of the cone is π −α, where α  1. 2 (ii) What is the nature of the singularity in pressure at the vertex of the cone (point A). (iii) Now suppose that the cone is truncated as shown in Figure 32.4. Without finding the new contact pressure distribution, find the new relationship between F and a. Note: Take the radius of the truncated end of the cone as b and refer the profile to the point A where the vertex would have been before truncation. (iv) Would you expect a singularity in contact pressure at B and if so of what form. (The process of determining p(r ) is algebraically tedious. Try to find a simpler way to answer this question.) 22

This solution was first given by K. L. Johnson, K. Kendall and A. D. Roberts (1971), Surface energy and the contact of elastic solids, Proceedings of the Royal Society of London, Vol.A324, pp. 301–313.

Problems

535

α Fig. 32.3 The conical punch.

α Fig. 32.4 The truncated conical punch.

32.7. A rigid flat punch has rounded edges, as shown in Figure 32.5. The punch is pressed into an elastic half-space by a force F. Assuming that the contact is frictionless, find the relation between F, the indentation depth u 0 and the radius a of the contact area.

Fig. 32.5 Flat punch with rounded corners.

32.8. An elastic half-space is indented by a smooth axisymmetric rigid punch with the power law profile Cr s , so the displacement in the contact area is u 0 (r ) = d − Cr s , where C, s are constants. Show that the indentation force F, the contact radius a and the indentation depth d are related by the equation F=

4μad (1 − ν)



s s+1

.

536

32 The Boundary-value Problem

32.9. The profile of a smooth axisymmetric frictionless rigid punch is described by the power law u 0 (r ) = An r 2n , where n is an integer. The punch is pressed into an elastic half-space by a force F. Find the indentation depth d, the radius of the contact area a and the contact pressure distribution p(r ). Check your results by comparison with the Hertz problem of §32.2.5 and give simplified expressions for the case of the fourth order punch u 0 (r ) = A2 r 4 . 32.10. For a smooth axisymmetric contact problem, the contact pressure must be continuous at r = a. Usually this condition is imposed explicitly and it serves to determine the radius a of the contact region. However, it can also be guaranteed by ‘choice of form’ as suggested in §32.2.6. Use this strategy to solve the Hertzian contact problem, for which u 0 (r ) = −r/R, where R is the radius of the indenter. You will need to find a suitable representation of the form ∂ϕ = φi , ∂z where φi is one of the four functions in equation (32.10), chosen so as to satisfy the homogeneous boundary condition (zero contact pressure in r > a). Then use Tables 32.1, 32.2 to set up and solve an Abel equation for the unknown function gi (t). In particular, find the contact pressure distribution p(r ) and the total load F needed to establish a circular contact area of radius a. 32.11. The flat end of a rigid cylindrical punch of radius a is pressed into the curved surface of an elastic cylinder of radius R a by a centric force F sufficient to ensure contact throughout the punch face. Use Copson’s formula (32.36–32.38) to determine the contact pressure distribution and hence find the minimum value of F required for full contact. 32.12. Solve Problem 32.11 by relating the non-axisymmetric component of ϕ(r, θ, z) to a corresponding axisymmetric harmonic function f 0 (r, z), as in §32.3.1. Notice that the required function varies with cos(2θ), so you will need to apply equation (25.32) twice recursively to relate f 2 (r, z) to f 0 (r, z). This will introduce two arbitrary constants in the boundary conditions for f 0 and these must be determined from the continuity conditions g1 (a) = g1 (a) = 0.

Chapter 33

The Penny-shaped Crack

As in the two-dimensional case, we shall find considerable similarities in the formulation and solution of contact and crack problems. In particular, we shall find that problems for the plane crack can be reduced to boundary-value problems which in the case of axisymmetry can be solved using the method of Green and Collins developed in §32.2.

33.1 The Penny-shaped Crack in Tension The simplest axisymmetric crack problem is that in which a state of uniform tension σzz = S in an infinite isotropic homogeneous solid is perturbed by a plane crack occupying the region 0 ≤r < a, z = 0. Thus, the crack has the shape of a circular disk. This geometry has come to be known as the penny-shaped crack1 . As in Chapter 13, we seek the solution in terms of an unperturbed uniform stress field σzz = S (constant) and a corrective solution which tends to zero at infinity. The boundary conditions on the corrective solution are therefore σzz = −S; σzr = 0 ; 0 ≤ r < a , z = 0 σzz , σr z , σrr , σθθ → 0 ; R → ∞ . The corrective solution corresponds to the problem in which the crack is opened by an internal pressure of magnitude S and the body is not loaded at infinity. The problem is symmetrical about the plane z = 0 and it follows that on that plane there can be no shear stress σzx , σzy and no normal displacement u z . We can therefore reduce the problem to a boundary-value problem for the half-space z > 0 defined by the boundary conditions 1

I. N. Sneddon (1946), The distribution of stress in the neighbourhood of a crack in an elastic solid, Proceedings of the Royal Society of London, Vol. A187, pp. 226–260.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 J. R. Barber, Elasticity, Solid Mechanics and Its Applications 172, https://doi.org/10.1007/978-3-031-15214-6_33

537

538

33 The Penny-shaped Crack

σzx = σzy = 0 ; all r, z = 0

(33.1)

σzz = −S ; 0 ≤ r < a, z = 0 u z = 0 ; r > a, z = 0 .

(33.2) (33.3)

Equation (33.1) is a global condition and can be satisfied identically2 by using Solution F of Table 22.3. The remaining conditions (33.2, 33.3) then define the mixed boundary-value problem ∂2ϕ = S ; 0 ≤ r < a, z = 0 (33.4) ∂z 2 ∂ϕ = 0 ; r > a, z = 0 . (33.5) ∂z This problem can be solved by the method of §32.2. We note from Table 32.1 that (33.5) can be satisfied identically by representing ϕ in the form of φ3 of (32.10) —  a i.e. ϕ= g(t)F(r, z, t)dt 0

and the remaining boundary condition (33.4) then reduces to  r 1 d tg(t)dt =S ; 0≤r a, z = 0 which is  a x 2d x ∂2ϕ 2S d σzz = − 2 = − √ ∂z πr dr 0 r2 − x2   a a 2S , sin−1 − √ =− π r r 2 − a2 using Table 32.1 and (33.7). This of course is the corrective solution. To find the corresponding stress in the original crack problem, we must superpose the uniform tensile stress σzz = S, with the result   a a 2S ; r > a, z = 0 . cos−1 + √ σzz = π r r 2 − a2 The stress-intensity factor is defined as  K I = lim+ σzz 2π(r − a) = 2S



r →a

a . π

(33.8)

We could have obtained this result directly from g(x), without computing the complete stress distribution. We have 1 d σzz = − r dr



a 0

xg(x)d x , √ r2 − x2

and integrating by parts before performing the differentiation as in §32.2.5, we obtain g(a) g(0) σzz = √ − − r r 2 − a2

 0

a

g  (x)d x . √ r2 − x2

(33.9)

Now, unless g  (x) is singular at x = a, the only singular term in (33.9) will be the first, and the stress-intensity factor is therefore   π . K I = lim+ σzz 2π(r − a) = g(a) r →a a In the present example, g(a) = 2Sa/π, leading to (33.8) as before. We also compute the displacement u z at 0 ≤r < a, z = 0+ , which is (1 − ν) (1 − ν) ∂ϕ = μ ∂z μ √ 2 2 2(1 − ν)S a − r = . πμ

u z (r, 0+ ) = −

 r

a

g(x)d x √ x2 − r2

540

33 The Penny-shaped Crack

Since the crack opens symmetrically on each face, it follows that the crack-opening displacement is √ 4(1 − ν)S a 2 − r 2 δ(r ) = u z (r, 0 ) − u z (r, 0 ) = . πμ +

−

33.2 Thermoelastic Problems In addition to acting as a stress concentration in an otherwise uniform stress field, a crack will obstruct the flow of heat and generate a perturbed temperature field in components of thermal machines. The simplest investigation of this effect assumes that the crack acts as a perfect insulator, so that the heat is forced to flow around it. For the penny-shaped crack3 , the appropriate thermal boundary conditions are then qz = −K

∂T = 0 ; 0 ≤ r < a, z = 0 ∂z → q0 ; R → ∞ ,

(33.10) (33.11)

where q0 is a constant defining the unperturbed heat flux. As in isothermal problems, we construct the temperature field as the sum of a uniform heat flux and a corrective solution, the boundary conditions for which become qz = −K

∂T = −q0 ; 0 ≤ r < a, z = 0 ∂z →0 ; R→∞.

This problem is antisymmetric about the plane z = 0 and hence, with a suitable choice of datum, the temperature must be zero in the region r > a, z = 0. We can therefore convert the heat conduction problem into a boundary-value problem for the half-space z > 0 with boundary conditions qz = −q0 ; 0 ≤ r < a T =0 ; r >a.

(33.12) (33.13)

The temperature field and a particular thermoelastic solution can be constructed using Williams’ solution [Solution T of Table 23.1], in terms of which (33.12, 33.13) define the mixed boundary-value problem

3

This problem was first solved by A. L. Florence and J. N. Goodier (1963), The linear thermoelastic problem of uniform heat flow disturbed by a penny-shaped insulated crack, International Journal of Engineering Science, Vol. 1, pp. 533–540.

33.2 Thermoelastic Problems

541

∂2χ μα(1 + ν)q0 ; 0≤r a. ∂z

(33.14) (33.15)

We require that the temperature and hence ∂χ/∂z be continuous at r = a, z = 0 and hence in view of the arguments of §32.2.6, we choose to satisfy (33.15) using the function φ3 — i.e.  a

χ=

g3 (t)F(r, z, t)dt .

0

Condition (33.14) then defines the Abel equation 1 d r dr



r 0

tg3 (t)dt μα(1 + ν)q0 = ; 0≤r 0 which are σzz = σzr = 0 ; 0 ≤ r < a, z = 0

(33.17)

σzz = 0 ; u r = 0 ; r > a, z = 0 .

(33.18)

Equations (33.17) state that the crack surfaces are traction free and (33.18) are symmetry conditions. Notice that taken together, these conditions imply that σzz = 0 throughout the plane z = 0. This is therefore a global condition and can be satisfied by the choice of form. It is already satisfied by the particular solution (see Table 23.1) and hence it must also be satisfied by the additional isothermal solution, for which we therefore use Solution G of Table 22.3. Notice however that we shall use the symbol φ for the potential in Solution G to avoid confusion with χ in Solution T. The complete solution is therefore obtained by superposing Solutions T and G, the surface values of σzr and u r being

542

33 The Penny-shaped Crack

σzr =

∂2φ ∂χ ∂φ + ; 2μu r = 2(1 − ν) . ∂r ∂z∂r ∂r

The mixed conditions in (33.17, 33.18) then require  ∂χ 1 r tg3 (t)dt ∂2φ =− = √ ∂r ∂z ∂r r 0 r2 − t2 μα(1 + ν)q0 r ; 0 ≤ r < a, z = 0 , = 2(1 − ν)K and

∂φ = 0 ; r > a, z = 0 , ∂r

(33.19)

(33.20)

where we used Table 32.1 and (33.16) to evaluate the right-hand side of (33.19). We can satisfy (33.20) by defining φ such that 

a

φ=

g1 (t)F(r, z, t)dt ,

(33.21)

0

with the auxiliary condition



a

g1 (t)dt = 0 ,

(33.22)

0

(see Table 32.1). The other boundary condition (33.19) can then be integrated to give μα(1 + ν)q0 r 2 ∂φ = + C ; 0 ≤ r < a, z = 0 , ∂z 4(1 − ν)K where C is an arbitrary constant of integration, and hence 

r 0

g1 (t)dt μα(1 + ν)q0 r 2 +C ; 0≤r a, z = 0, which is

Problems

543

 a  ∂χ d g1 (t)dt ∂2φ 1 a tg3 (t)dt + = σzr (r, 0) = − √ √ ∂r ∂z ∂r dr 0 r 0 r2 − t2 r2 − t2 3 2μα(1 + ν)q0 a =− , √ 3π(1 − ν)K r r 2 − a 2 from Table 32.1 and equations (33.16, 33.24, 33.25). Notice that the antisymmetry of the problem ensures that the crack tip is loaded in shear only — we argued earlier that σzz would be zero throughout the plane z = 0. It also follows from a similar argument that the crack remains closed — there is a displacement u z at the crack plane, but it is the same on both sides of the crack. The stress-intensity factor is of mode II (shear) form and is  2μα(1 + ν)q0 a 3/2 . K II = lim+ σzr 2π(r − a) = − √ r →a 3 π(1 − ν)K

Problems 33.1. An infinite homogeneous body contains an axisymmetric external crack — i.e. the crack extends over the region r > a, z = 0. Alternatively, the external crack can be considered as two half-spaces bonded together over the region 0 ≤r < a, z = 0. The body is loaded in tension, such that the total tensile force transmitted is F. Find an expression for the stress field and in particular determine the mode I stressintensity factor K I . 33.2. An infinite homogeneous body containing a penny-shaped crack is subjected to torsional loading σzθ = Cr, R → ∞, where C is a constant4 . Use Solution E (Table 22.1) to formulate the problem and solve the resulting boundary-value problem using the methods of §32.2. 33.3. The axisymmetric external crack of radius a (see Problem 33.1) is unloaded, but is subjected to the steady-state thermal conditions T (r, z) → T1 ; z → ∞ → T2 ; z → −∞ qz (r, 0) = 0 ; r > a . In other words the extremities of the body are maintained at different temperatures T1 , T2 causing heat to flow through the ligament z = 0, 0 ≤r < a and the crack faces are insulated. Find the mode II stress-intensity factor K II . 4

This is known as the Reissner-Sagoci problem.

544

33 The Penny-shaped Crack

33.4. The antisymmetry of the thermoelastic penny-shaped crack problem of §33.2 implies that there is no crack-opening displacement and hence the assumption of insulation (33.10) is arguably rather unrealistic. A more realistic assumption might be that the heat flux across the crack faces is proportional to the local temperature difference — i.e.   qz (r, 0) = h T (r, 0− ) − T (r, 0+ ) ; 0 ≤ r < a . Use the methods of Chapter 32 to formulate the heat conduction problem. In particular, express the temperature in terms of one of the forms (32.10) and find the Abel integral equation which must be satisfied by the function g3 (t). Do not attempt to solve this equation. 33.5. A long rectangular beam defined by the cross section −c < x < c, −d < y < d transmits a bending moment M about the positive x-axis. The beam contains a small penny-shaped crack of radius a in the cross-sectional plane z = 0 with its centre at the point (0, b, 0) — i.e. the crack surface is defined by 0 ≤ x 2 +(y −b)2 < a, z = 0. Assuming that the crack opens completely and that c, d, d −b are all large compared with a, find the variation of K I around the crack edge and hence determine the minimum value of b for the open-crack assumption to be valid. Find also the crack-opening displacement as a function of position within the crack. Notice that the corrective problem involves both axisymmetric and nonaxisymmetric terms. For the latter, you will need to apply the methods introduced in §32.3. 33.6. The otherwise plane surface of an elastic half-space contains an axisymmetric dimple whose depth below the plane can be described by the function  2 r2 ; 0≤r b, where 0 < b < a, find the relation between b, σ0 , Δγ and the elastic properties of the materials. Assume that the contact is frictionless. Plot an appropriate dimensionless measure of σ0 as a function of b/a for some representative values of h 0 and comment on the implications of these plots for the stability of the corresponding contact configurations.

Chapter 34

Hertzian Contact

In this chapter, we shall consider the frictionless contact of two elastic bodies with profiles that can be described by general quadratic functions. This problem was first solved by Hertz1 and has numerous practical applications. Special cases where the bodies are two-dimensional or axisymmetric were already considered in §12.7.1 and §32.2.5 respectively.

Fig. 34.1 Definition of profiles through gap functions.

Figure 34.1 shows two bodies with fairly general profiles that are placed in contact, but not loaded, so that contact occurs only at a single point O. The two surfaces must share a common tangent plane at O, since any other assumption would imply local interpenetration of material. We therefore define Cartesian coördinates such that x, y lie in this plane and z is the normal pointing into the lower body 1.

Supplementary Information The online version contains supplementary material available at https://doi.org/10.1007/978-3-031-15214-6_34. 1

H. Hertz (1882), Über die Berührung fester elastischer Körper (On the contact of stiff elastic solids), Journal für die reine und angewandte Mathematik, Vol 92, pp. 156–171. (in German).

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 J. R. Barber, Elasticity, Solid Mechanics and Its Applications 172, https://doi.org/10.1007/978-3-031-15214-6_34

545

546

34 Hertzian Contact

We then define the profiles of the two bodies by two gap functions g1 (x, y), g2 (x, y) as shown in Figure 34.1, such that gi (x, y), i = 1, 2 defines the gap if body i is placed in contact with a rigid plane. The total gap as defined in Figure 12.12 is then given by g0 (x, y) = g1 (x, y) + g2 (x, y) , and if both bodies have quadratic gap functions, g0 will also be a quadratic function of x and y. Furthermore, since we take the initial contact point as origin, and since the gap functions are assumed to be twice differentiable (i.e. no discontinuous changes of slope), we must have g0 (0, 0) = 0 ;

∂g0 ∂g0 (0, 0) = 0 ; (0, 0) = 0 . ∂x ∂y

Thus, the second derivatives of the gap function  2  ∂ g0 ∂ 2 g0 ∂ 2 g0 ∂ 2 g0 , , , ∂x 2 ∂x∂ y ∂ y∂x ∂ y2

(34.1)

can be considered as defining the first non-zero terms in a Taylor series expansion of a general function g0 (x, y), and since the contact region is generally expected to be small, the Hertzian theory provides a good approximation to the contact problem for fairly general smooth convex surfaces — not just those where the contacting surfaces are strictly quadratic. The second derivatives (34.1) define a second order Cartesian tensor, so there is no loss of generality in taking the principal directions as the coördinate directions, in which case the most general quadratic gap function can be written g0 (x, y) = Ax 2 + By 2 .

(34.2)

We recover the two-dimensional case of §12.7.1 if B = 0, and the axisymmetric case of §32.2.5 if A = B. For all other values, we shall show that the contact area A is an ellipse which we shall define as x2 y2 + 2 0 produces a normal surface displacement u z (r, 0) =

F(1 − ν) , 2πμr

(34.5)

from equation (24.6). It follows that an equal and opposite compressive force pair at the origin will produce F(1 − ν1 ) F(1 − ν2 )   + 2 2 2πμ1 x + y 2πμ2 x 2 + y 2 F  = , ∗ π E x 2 + y2

(2) u (1) z (x, y, 0) − u z (x, y, 0) =

(34.6)

∗

where the composite contact modulus2 E is defined as 1 (1 − ν12 ) (1 − ν22 ) (1 − ν1 ) (1 − ν2 ) + = + . ∗ = E1 E2 2μ1 2μ2 E Using (34.6) as a Green’s function, we can then write the contact condition (34.3) as an integral equation 1 πE∗ 2

 A

p(ξ, η)dξdη  = Δ − Ax 2 − By 2 ; (x, y) ∈ A . (x − ξ)2 + (y − η)2 ∗

Notice that for plane strain [κ = (3 − 4ν)], the constant A of equation (12.42) is equal to 8/E .

548

34 Hertzian Contact

34.1.1 Field-point integration In evaluating this integral, it is advantageous to move the origin to a field point P(x, y) and revert to polar coördinates (r, θ) centred on P, since then the factor ‘r ’ in the elemental area r dθdr cancels with that in the Green’s function (34.5). We obtain  π/2  S2 1 p(r, θ)dr dθ = Δ − Ax 2 − By 2 , (34.7) π E ∗ −π/2 S1 where the points S1 , S2 are defined in Figure 34.2. Notice that we include contributions from the entire line S1 S2 in the inner integral, so that r should be interpreted as negative in the segment S1 P. Fig. 34.2 Elliptical contact area.

The coördinates of a general point Q(ξ, η) are then defined through ξ = x + r cos θ ; η = y + r sin θ , so the pressure distribution (34.4) is 

(x + r cos θ)2 (y + r sin θ)2 − p(r, θ) = p0 1 − a2 b2  2 = p0 C0 − C1 (θ)r − C2 (θ)r ,

(34.8)

where   x 2 y2 x cos θ y sin θ cos2 θ sin2 θ ; C − ; C (θ) = 2 + (θ) = + 2 . 1 2 a 2 b2 a2 b2 a2 b (34.9) Also, the limits r1 , r2 for the inner integral correspond to the two roots of the quadratic equation C0 =1 −

34.1 Elastic Deformation

549

C0 − C1 (θ)r − C2 (θ)r 2 = 0 , since p(r, θ) = 0 at S1 and S2 , from (34.4). If (34.8) is substituted into (34.7), the inner integral can be evaluated in terms of elementary functions as 

S2

 p(r, θ)dr = p0

S1

r2

r1

π p0 = √ 2 C2 and hence p0 2E ∗



 C0 − C1 (θ)r − C2 (θ)r 2 dr   C2 C0 + 1 , 4C2

  dθ C12 C0 + √ = Δ − Ax 2 − By 2 . 4C C2 2 −π/2 π/2

In view of the definitions (34.9), the left-hand side of this equation will clearly evaluate to an even quadratic function of x and y, so it can be satisfied exactly if a, b and p0 are chosen appropriately. Equating coefficients of x 2 , y 2 and constants respectively, we obtain  sin2 θdθ p0 ab π/2 A= ∗ E (b2 cos2 θ + a 2 sin2 θ)3/2 0  π/2 p0 ab cos2 θdθ B= ∗ E (b2 cos2 θ + a 2 sin2 θ)3/2 0  π/2 p0 ab dθ  Δ= . ∗ E 0 b2 cos2 θ + a 2 sin2 θ

(34.10) (34.11) (34.12)

Greenwood3 shows that these results can be expressed in terms of Carlson elliptic integrals using the change of variable t = cot 2 θ. We obtain   a2 p0 a R D 0, 1; 2 ; A= b 3E ∗ b2 Δ=

  p0 b b2 B= R D 0, 1; 2 ; a 3E ∗ a 2

    p0 a a2 p0 b b2 0, 1, = 0, 1, , R R F F b2 a2 E∗ E∗

(34.13)

(34.14)

where

3 J. A. Greenwood (2018), Hertz theory and Carlson elliptic integrals, Journal of the Mechanics and Physics of Solids, Vol. 119, pp. 240–249. Notice that Greenwood uses a modified definition of R D without the factor of 3, which he then denotes by R ⊗ D.

550

34 Hertzian Contact

 dt 1 ∞ √ 2 0 (t + p)(t + q)(t + r )  dt 3 ∞ . R D ( p, q; r ) = √ 2 0 (t + r ) (t + p)(t + q)(t + r ) R F ( p, q, r ) =

The expressions (34.10–34.12) can also be defined in terms of the more familiar Legendre elliptic integrals 

π/2

K (k) =

√ 0



dθ 1−

k2

cos2

θ

;

π/2

E(k) =

 1 − k 2 cos2 θdθ

0

as p0 b A = ∗ 2 2 [K (k) − E(k)] ; E a k

p0 b B= ∗ 2 2 E a k

p0 bK (k) , Δ= E∗



E(k) − K (k) (1 − k 2 ) 

where

k=

1−

 (34.15)

b2 a2

is the eccentricity of the contact ellipse. The definitions of Legendre’s integrals require that the axes be oriented such that b ≤ a and hence B ≥ A, whereas the original expressions (34.10–34.12) and the Carlson integral expressions (34.13, 34.14) are clearly symmetric with respect to the axis directions, and hence do not require this preliminary step.

34.2 Solution Procedure In a typical technical problem, the coefficients A and B will be determined from the shape of the contacting bodies, and either the rigid-body displacement Δ, or the applied force  2π p0 ab (34.16) p(x, y)d xd y = F= 3 A will be prescribed. Equations (34.13) and either (34.14) or (34.16) then define three equations for the unknowns a, b, p0 . From equations (34.15) or (34.13), we obtain 1 B = A [K (k) − E(k)]



  3 E(k) b R D (0, 1; b2 /a 2 ) − K (k) = , 2 (1 − k ) a R D (0, 1; a 2 /b2 )

(34.17)

which shows that k, or equivalently the ratio b/a, is a unique function of B/A. Thus the shape of the ellipse remains unchanged as the load or the rigid-body displacement is increased, and it can be found by solving this non-linear equation. This relationship

34.2 Solution Procedure

Fig. 34.3 Relation between

551

√

A/B and b/a.

√ is shown in Figure 34.3, where we note that A/B represents the ratio of the semiaxes of the elliptical contours of the undeformed composite surface g0 (x, y) from √ equation (34.2). We note that b/a is always less than A/B and hence the contact ellipse has a larger eccentricity than that of the undeformed contours.

34.2.1 Axisymmetric bodies In many applications, the contacting bodies will be axisymmetric — typically cylinders, spheres or cones — and the individual gap functions g1 (x, y), g2 (x, y) are then of particularly simple form. For example, for a sphere of radius R we have g(x, y) =

x 2 + y2 , 2R

and for a cylinder of radius R with its axis aligned with the x-axis, g(x, y) =

y2 . 2R

However, if the gap functions g1 , g2 thereby determined are in different Cartesian coördinate systems as in Problem 34.1, it will be necessary to use the tensor

552

34 Hertzian Contact

transformation rules (1.16) or equivalently the Mohr’s circle equations (1.9–1.11) to rotate them into a common system and then to determine the principal values A, B in equation (34.2).

Example Figure 34.4 shows a cross-sectional view of a railway wheel in contact with a rail. The radius of the wheel at the contact point is 0.5 m and the contact surface is a cone of angle 5o as shown. The rail contact surface can be approximated as a cylinder of radius 100 mm. Both bodies are made of steel for which E = 210 GPa, ν = 0.3. If the wheel transmits a vertical force of 10 kN, find the maximum contact pressure p0 and the dimensions of the contact area. Fig. 34.4 Wheel-rail contact

Both materials are of steel, so ∗

E =

210 E = = 115 GPa . 2 2(1 − ν ) 2 × 0.91

The contact force must be F = 10, 000/ cos(5o ) = 10, 038 N in order to have a vertical component of 10 kN. The local radius of the conical surface is 500 = 502 mm , cos(5o )

Problems

553

so if we define the x-direction as parallel with the rail, we have y2 x2 ; g2 (x, y) = g1 (x, y) = 200 1004 and g0 (x, y) = g1 (x, y) + g2 (x, y) = Ax 2 + By 2 with A = 0.000996 mm−1 ;

B = 0.005 mm−1 .

Noting that B > A, we can solve equation (34.17) for k, obtaining b  = 1 − k 2 = 0.345 . k = 0.9386 or a Maple and Mathematica codes for this solution are available in the files ‘Hertz’. Alternatively, we could use the above values of A, B and estimate b/a from Figure 34.3. Eliminating p0 between equations (34.16) and (34.15)1 , we obtain     3 × 10, 038 × [K (0.939) − E(0.939)] 1/3 3F[K (k) − E(k)] 1/3 = a= 2π × 0.000996 × 115 × 103 × 0.9392 2π AE ∗ k 2 = 4.10 mm so b = 0.345 × 4.10 = 1.41 mm and p0 =

3 × 10, 038 3F = = 828 MPa , 2πab 2π × 4.10 × 1.41

noting that 1 MPa = 1 N/mm2 .

Problems 34.1. Two identical cylinders of radius R are pressed together by a force F. If the axes of the two cylinders are inclined at an angle of 45o , find the ratio of the semi-axes ∗ b/a and the maximum contact pressure p0 in terms of F, R and E . 34.2. Find a relation between the force F and the indentation depth Δ that depends ∗ only on the shape of the contacting bodies and E . Hence show that the incremental stiffness dF = C F 1/3 dΔ where C is a constant.

554

34 Hertzian Contact

34.3. An incompressible elastic half-space with modulus μ = 400 MPa has a sinusoidal profile defined by g(x) = −2 cos(0.1x) , where g and x are in mm. A rigid sphere of radius 40 mm is pressed into the trough of the sinusoid by a force F. Find the eccentricity of the resulting elliptical contact and estimate the value of F beyond which the Hertzian theory can be expected to be significantly in error. 34.4. For a given Hertzian contact problem, the applied force F, the contact pressure p(x, y) and the displacements u z (x, y) are related by equations (34.16, 34.4, 34.3, 34.13, 34.14). Find the corresponding quantities for a force F + ΔF applied to the same elastic bodies and construct the difference between these two solutions — e.g. p(x, y, F + ΔF) − p(x, y, F) and the corresponding incremental displacements. By taking the limit as ΔF → 0, find the contact pressure distribution and the indentation depth for a flat rigid punch of elliptical cross section with semi-axes a, b. 34.5. Figure 34.5 shows a ball bearing with an inner race of radius 36 mm, an outer race of radius 48 mm and 16 balls of 12 mm diameter. Both inner and outer race surfaces have a (concave) groove of radius 12 mm.

Fig. 34.5 A ball bearing

Find the eccentricity k and the semi-axis ratio a/b of the contact area between the balls and the inner race and also for the contact area at the outer race.

Chapter 35

The Interface Crack

The problem of a crack at the interface between dissimilar elastic media is of considerable contemporary importance in Elasticity, because of its relevance to the problem of debonding of composite materials and structures. Figure 35.1 shows the case where such a crack occurs at the plane interface between two elastic half-spaces. We shall use the suffices 1,2 to distinguish the stress functions and mechanical properties for the lower and upper half-spaces, z > 0, z < 0, respectively.

μ ν μ ν Fig. 35.1 The plane interface crack.

The difference in material properties destroys the symmetry that we exploited in the previous chapter and we therefore anticipate shear stresses as well as normal stresses at the interface, even when the far field loading is a state of uniform tension. However, we shall show that we can still use the same methods to reduce the problem to a mixed boundary-value problem for the half-space.

35.1 The Uncracked Interface The presence of the interface causes the problem to be non-trivial, even if there is no crack and the two bodies are perfectly bonded. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 J. R. Barber, Elasticity, Solid Mechanics and Its Applications 172, https://doi.org/10.1007/978-3-031-15214-6_35

555

556

35 The Interface Crack

Consider the tensile loading of the composite bar of Figure 35.2(a).

Fig. 35.2 The composite bar in tension.

If we assume a state of uniform uniaxial tension σzz = S exists throughout the bar, the Poisson’s ratio strains, ex x = e yy = −νi S/E i will imply an inadmissible discontinuity in displacements u x , u y at the interface, unless ν1 /E 1 = ν2 /E 2 . In all other cases, a locally non-uniform stress field will be developed involving shear stresses on the interface and the bar will deform as shown in Figure 35.2(b). Indeed, for many material combinations, there will be a stress singularity at the edges A, B, which can be thought of locally as two bonded orthogonal wedges1 and analyzed by the method of §11.2. We shall not pursue the perfectly-bonded problem here — a solution for the corresponding two-dimensional problem is given by Bogy2 — but we note that as long as the solution for the uncracked interface is known, the corresponding solution when a crack is present can be obtained as in §§13.3.2, 33.1, by superposing a corrective solution in which tractions are imposed on the crack face, equal and opposite to those transmitted in the uncracked state, and the distant boundaries of the body are traction free. Furthermore, if the crack is small compared with the other linear dimensions of the body — notably the distance from the crack to the free boundary — it can be conceived as occurring at the interface between two bonded half-spaces, as in Figure 35.1.

1

D. B. Bogy (1968), Edge-bonded dissimilar orthogonal elastic wedges under normal and shear loading, ASME Journal of Applied Mechanics, Vol. 35, pp. 460–466. 2 D. B. Bogy (1975), The plane solution for joined dissimilar elastic semistrips under tension, ASME Journal of Applied Mechanics., Vol. 42, pp. 93–98.

35.2 The Corrective Solution

557

35.2 The Corrective Solution If we assume that the loading conditions are such as to cause the crack to open fully and therefore to be traction free, the corrective solution will be defined by the boundary conditions + (2) − σx(1) z (x, y, 0 ) = σx z (x, y, 0 ) = −S1 (x, y) + (2) − σ (1) yz (x, y, 0 ) = σ yz (x, y, 0 ) = −S2 (x, y)

(35.1)

(1) (2) (x, y, 0+ ) = σzz (x, y, 0− ) = −S3 (x, y) , σzz

for (x, y) ∈ A and (i) (i) (i) (i) σx(i)x , σ (i) yy , σzz , σx y , σ yz , σzx → 0 ; R → ∞ ,

(35.2)

where A is the region of the interfacial plane z = 0 occupied by the crack and S1 , S2 , S3 are the tractions that would be transmitted across the interface in the absence of the crack. We shall seek a potential function solution of the problem in the half-space, so we supplement (35.1–35.2) by the continuity and equilibrium conditions + (2) − σx(1) z (x, y, 0 ) = σx z (x, y, 0 ) + σ (1) yz (x, y, 0 ) (1) (x, y, 0+ ) σzz u (1) x (x, y, 0+) (1) u y (x, y, 0+) u (1) z (x, y, 0+)

= = = = =

− σ (2) yz (x, y, 0 ) (2) σzz (x, y, 0− ) u (2) x (x, y, 0−) (2) u y (x, y, 0−) u (2) z (x, y, 0−)

(35.3) (35.4) (35.5) (35.6) (35.7) ,

(35.8)

for (x, y) ∈ A¯ — i.e. the uncracked part of the interface3 . It is clear from (35.1) that (35.3–35.5) apply throughout the interface and are therefore global conditions. We can exploit this fact by expressing the fields in the two bonded half-spaces by separate sets of stress functions and then developing simple symmetric relations between the sets.

35.2.1 Global conditions The global conditions involve the three stress components σzx , σzy , σzz at the surface z = 0, so it is convenient to choose a formulation in which these components take 3

Notice that only those stress components that act on the interface (and hence have a z-suffix) are continuous across the interface, because of the equilibrium requirement. The remaining three components will generally be discontinuous because of (35.6–35.8) and the dissimilar elastic properties.

558

35 The Interface Crack

simple forms at the surface and are as far as possible uncoupled. This can be achieved by superposing Solutions E,F,G of Chapter 22, for which ∂ 2 Ψi ∂ 2 χi + ∂ y∂z ∂x∂z 2 ∂ 2 χi Ψ ∂ i (i) + (x, y, 0) = − σzy ∂x∂z ∂ y∂z 2 ϕ ∂ i (i) σzz =− 2 , ∂z

(i) (x, y, 0) = σzx

(35.9) (35.10) (35.11)

from Tables 22.1, 22.3, where the index i takes the value 1,2 for bodies 1,2 respectively. The global condition (35.5) now reduces to ∂ 2 ϕ2 ∂ 2 ϕ1 + (x, y, 0 ) = (x, y, 0− ) ∂z 2 ∂z 2 and, remembering that ϕ1 is defined only in z > 0 and ϕ2 only in z < 0, we can satisfy it by imposing the symmetry relation ϕ1 (x, y, z) = ϕ2 (x, y, −z) ≡ ϕ(x, y, z)

(35.12)

throughout the half-spaces. In the same way4 , the two remaining global conditions can be satisfied by demanding Ψ1 (x, y, z) = −Ψ2 (x, y, −z) ≡ Ψ (x, y, z) χ1 (x, y, z) = −χ2 (x, y, −z) ≡ χ(x, y, z) ,

(35.13) (35.14)

where the negative sign in (35.13, 35.14) arises from the fact that symmetry of the derivatives ∂Ψ/∂z, ∂χ/∂z implies antisymmetry of Ψ, χ.

35.2.2 Mixed conditions We can now use the remaining boundary conditions (35.1) and (35.6–35.8) to define a mixed boundary-value problem for the three potentials ϕ, Ψ, χ in the half-space z > 0. For example, using the expressions from Tables 22.1, 22.3 and the definitions (35.12–35.14), the boundary condition (35.6) reduces to 4

This idea can also be extended to steady-state thermoelastic problems by adding in Solution P of Table 23.1, which leaves the expressions (35.9–35.11) for the interface stresses unchanged and requires a further symmetry relation between ψ1 , ψ2 to satisfy the global condition of continuity of (1) (2) heat flux, qz (x, y, 0+ ) = qz (x, y, 0− ). See Problem 35.2.

35.2 The Corrective Solution

559

1 ∂Ψ (1 − 2ν2 ) ∂ϕ (1 − ν2 ) ∂χ 1 ∂Ψ (1 − 2ν1 ) ∂ϕ (1 − ν1 ) ∂χ + + =− + − μ1 ∂ y 2μ1 ∂x μ1 ∂x μ2 ∂ y 2μ2 ∂x μ2 ∂x ¯ z = 0. in A, Rearranging the terms and dividing by the non-zero factor [(1−ν1 )/μ1 +(1− ν2 )/μ2 ] (which is A/4 in the terminology of equation (12.42) for plane strain), we obtain ∂Ψ ∂ϕ ∂χ γ +β + = 0 ; in A¯ , (35.15) ∂y ∂x ∂x where β is Dundurs’ constant (see §§4.3.3, 12.7) and  γ=

1 1 + μ1 μ2



(1 − ν1 ) (1 − ν2 ) + μ1 μ2

 .

A similar procedure applied to the boundary conditions (35.7, 35.8) yields −γ

∂Ψ ∂ϕ ∂χ +β + =0 ∂x ∂y ∂y ∂χ ∂ϕ +β =0, ∂z ∂z

(35.16) (35.17)

¯ in A. In addition to (35.15, 35.16, 35.17), three further conditions are obtained from (35.1), using (35.9–35.11) and (35.12–35.14). These are ∂2Ψ ∂2χ + = −S1 (x, y) ∂ y∂z ∂x∂z ∂2χ ∂2Ψ + = −S2 (x, y) − ∂x∂z ∂ y∂z ∂2ϕ − 2 = −S3 (x, y) , ∂z

(35.18) (35.19) (35.20)

in A. The six conditions (35.15, 35.16–35.20) define a well-posed boundary-value problem for the three potentials ϕ, Ψ, χ, when supplemented by the requirements that (i) displacements should be continuous at the crack boundary (between A and ¯ and (ii) the stress and displacement fields should decay as R → ∞ in accordance A) with (35.2). The two boundary conditions (35.15, 35.16) involve only two independent functions Ψ and (βϕ+χ) and hence we can eliminate either by differentiation with the result

560

35 The Interface Crack

∂2Ψ ∂2Ψ ∂2Ψ + = − =0 ∂x 2 ∂ y2 ∂z 2   2 ∂2ϕ ∂2χ ∂ ∂2 (βϕ + χ) = −β + − 2 =0 ∂x 2 ∂ y2 ∂z 2 ∂z

(35.21) (35.22)

¯ where we have used the fact that the potential functions are all harmonic to in A, express the boundary conditions in terms of a single derivative normal to the plane. In the same way, (35.18, 35.19) yield ∂3Ψ ∂ S1 ∂ S2 = − 3 ∂z ∂y ∂x ∂ S2 ∂ S1 ∂3χ + , = ∂z 3 ∂x ∂y

(35.23) (35.24)

in A. Equations (35.21, 35.23) now define a two-part boundary-value problem for Ψ , similar to those solved in Chapter 32 for the axisymmetric geometry. It should be noted however that it is not necessary, nor is it generally possible, to satisfy the requirement of continuity of in-plane displacement u x , u y at the crack boundary in the separate contributions from the potential functions ϕ, Ψ, χ, as long as the superposed fields satisfy this requirement. Thus, the homogeneous problem for Ψ obtained by setting the right-hand side of (35.23) to zero has non-trivial solutions which are physically unacceptable in isolation, since they correspond to displacement fields which are discontinuous at the crack boundary, but they are an essential component of the solution of more general problems, where they serve to cancel similar discontinuities resulting from the fields due to ϕ, χ. This problem also arises for the crack in a homogeneous medium with shear loading5 and in non-axisymmetric thermoelastic problems6 The remaining boundary conditions define a coupled two-part problem for ϕ, χ. As in the two-dimensional contact problems of Chapter 12, the coupling is proportional to the Dundurs’ constant β. If the material properties are such that β = 0 (see §12.7), the problems of normal and shear loading are independent and (for example) the crack in a shear field has no tendency to open or close.

5

See for example J. R. Barber (1975), The penny-shaped crack in shear and related contact problems, International Journal of Engineering Science, Vol. 13, pp. 815–832. 6 L. Rubenfeld (1970), Non-axisymmetric thermoelastic stress distribution in a solid containing an external crack, International Journal of Engineering Science, Vol. 8, pp. 499–509, J. R. Barber (1975), Steady-state thermal stresses in an elastic solid containing an insulated penny-shaped crack, Journal of Strain Analysis, Vol. 10, pp. 19–24.

35.3 The Penny-shaped Crack in Tension

561

35.3 The Penny-shaped Crack in Tension We now consider the special case of the circular crack 0 ≤r < a, subjected to uniform tensile loading S3 = S, S1 = S2 = 0. The boundary-value problem for ϕ, χ is then defined by the conditions7 ∂3χ ∂z 3 ∂2ϕ ∂z 2 ∂ϕ ∂χ +β ∂z ∂z ∂2ϕ ∂2χ β 2 + 2 ∂z ∂z

= 0 ; z = 0, 0 ≤ r < a

(35.25)

= S ; z = 0, 0 ≤ r < a

(35.26)

= 0 ; z = 0, r > a

(35.27)

= 0 ; z = 0, r > a .

(35.28)

From Table 32.1, we see that we can satisfy (35.27, 35.28) by writing8 βϕ + χ = φ1 , ϕ + βχ = φ3 — i.e. ϕ=

φ1 − βφ3 φ3 − βφ1 , χ= , (1 − β 2 ) (1 − β 2 )

(35.29)

where φ1 , φ3 are defined by equation (32.10). The remaining boundary conditions then define two coupled integral equations of Abel-type for the unknown functions g1 (t), g3 (t) in the range 0 ≤ t < a. To develop these equations, we first note that the condition ∇ 2 χ = 0 and the fact that the solution is axisymmetric enable us to write (35.25) in the form 1 d d ∂χ r (r, 0) = 0 ; z = 0, 0 ≤ r < a . r dr dr ∂z This can be integrated within the plane z = 0 to give ∂χ = C ; z = 0, 0 ≤ r < a , ∂z 7

(35.30)

This problem was first considered by V. I. Mossakovskii and M. T. Rybka (1964), Generalization of the Griffith-Sneddon criterion for the case of a non-homogeneous body, Journal of Applied Mathematics and Mechanics, Vol. 28, pp. 1277–1286, and by F. Erdogan (1965), Stress distribution on bonded dissimilar materials containing circular or ring-shaped cavities, ASME Journal of Applied Mechanics, Vol. 32, pp. 829–836. A formulation for the penny-shaped crack with more general loading was given by J. R. Willis (1972), The penny-shaped crack on an interface,Quarterly Journal of Mechanics and Applied Mathematics, Vol. 25, pp. 367–382. 8 This method would fail for the special case β = ±1, but materials with positive Poisson’s ratio are restricted by energy considerations to values in the range − 21 ≤ β ≤ 21 .

562

35 The Interface Crack

where C is a constant which will ultimately be chosen to ensure continuity of displacements at r = a, and we have eliminated a logarithmic term to preserve continuity of displacements at the origin. Table 32.1 and (35.29, 35.30) now enable us to write 

g1 (t)dt +β √ r2 − t2

r 0



a

r

g3 (t)dt = (1 − β 2 )C ; 0 ≤ r < a , √ t2 − r2

(35.31)

whilst (35.26, 35.29) and Table 32.1 give 1 d r dr



r 0

tg3 (t)dt β d − √ 2 2 r dr r −t



tg1 (t)dt = (1 − β 2 )S ; 0 ≤ r < a , (35.32) √ t2 − r2

a

r

which can be integrated to give  0

r

tg3 (t)dt −β √ r2 − t2



a

r

tg1 (t)dt (1 − β 2 )Sr 2 = +B ; 0≤r a √ 2 (1 − β ) dr 0 r2 − t2  a d tg3 (t)dt 1 σzz = − ; z = 0, r > a . √ 2 (1 − β )r dr 0 r2 − t2

σzr =

(35.40) (35.41)

35.3.2 Oscillatory singularities An asymptotic analysis of the integrals (35.40, 35.41) at r = a +s, s a, shows that they have the form  s − 21 +ı + O(1) 2a 1   s − 2  s 

s  + ı sin ln + O(1) , cos ln = SK 2a 2a 2a (35.42)

σzz + ıσzr = S K

where

  1+β 1 ln = 2π 1−β

and Kassir and Bregman12 have found the complex constant K to be 1 Γ (2 + ı ) . K =√ π Γ ( 21 + ı ) Equation (35.42) shows that the stresses are square-root singular at the crack tip, but that they also oscillate with increasing frequency as r approaches a and ln(s/2a) → −∞. This behaviour is common to all interface crack problems for which β = 0 and can be predicted from an asymptotic analysis similar to that of §11.2.2, from which a complex leading singular eigenvalue λ = 21 ±ı is obtained.

35.4 The Contact Solution A more disturbing feature of this behaviour is that the crack opening displacement, (2) − (u (1) z (r, 0)−u z (r, 0)), also oscillates as r → a and hence there are infinitely many regions of interpenetration of material near the crack tip.

12

M. K. Kassir and A. M. Bregman (1972), The stress-intensity factor for a penny-shaped crack between two dissimilar materials, ASME Journal of Applied Mechanics, Vol. 39, pp. 308–310.

35.4 The Contact Solution

565

This difficulty was first resolved by Comninou13 for the case of the plane crack, by relaxing the superficially plausible assumption that the crack will be fully open and hence have traction-free faces in a tensile field, using instead the more rigorous requirements of frictionless unilateral contact — i.e. permitting contact to occur, the extent of the contact zones (if any) to be determined by the inequalities (31.6, 31.7). She found that a very small contact zone is established adjacent to each crack tip, the extent of which is of the same order of magnitude as the zone in which interpenetration is predicted for the original ‘open crack’ solution. At the closed crack tip, only the shear stresses in the bonded region are singular, leading to a ‘mode II’ (shear) stress-intensity factor, by analogy with equation (13.21). However, the contact stresses show a proportional compressive singularity in r → a − . Both singularities are of the usual square-root form, without the oscillatory character of equations (35.42). Solutions have since been found for other interface crack problems with contact zones, including the plane crack in a combined tensile and shear field14 . These original solutions of the interface crack problem used numerical methods to solve the resulting integral equation, but more recently an analytical solution has been found by Gautesen and Dundurs15 . The problem of §35.3 — the penny-shaped crack in a tensile field — was solved in a unilateral contact formulation by Keer et al.16 . They found a small annulus of contact b 0, respectively. Show that the resulting integral equations can be combined as in §35.3.1, but do not attempt to solve the resulting equation. 35.2. The interface penny-shaped crack of Figure 35.1 is unloaded, but the crack acts as an insulated obstruction to an otherwise uniform heat flux qz = q0 . Represent the temperature and displacement fields by appropriate potential functions using functions of the form (32.10), but do not attempt to solve the resulting integral equations. 35.3. The end z = 0 of the semi-infinite cylinder 0 ≤r < a, z > 0 is protected by a thermal barrier coating of thickness t. The bodies are stress free at the assembly

20

J. R. Rice (1988), Elastic fracture mechanics concepts for interfacial cracks, ASME Journal of Applied Mechanics, Vol. 55, pp. 98–103.

568

35 The Interface Crack

temperature T0 , but we wish to investigate the stress state at a uniform higher temperature T , in particular to assess the risk of delamination. For simplicity, suppose that both materials have the same elastic properties E, ν, but that the thermal expansion coefficients of the cylinder and the coating are α1 , α2 respectively, with α1 > α2 . Show that no tractions will be transmitted across the interface if an appropriate normal traction is applied to the curved surfaces of the coating r = a, −t < z < 0. If this traction is not applied, will the stress state at r = a, z = 0 involve a singularity, and if so, of what form. 35.4. It is proposed to formulate the two-dimensional interface crack problem using a distribution of dislocations, as in §13.3. Find the Airy stress function for a climb dislocation at the interface — i.e. the solution of the problem u y (x, 0+ ) − u y (x, 0− ) = B y H (x) ;

u x (x, 0+ ) − u x (x, 0− ) = 0 ,

if the elastic constants are μ1 , κ1 for y < 0 and μ2 , κ2 for y > 0. 35.5. If β = 0, equation (35.42) defines an oscillatory singularity at the tip of an open interface crack. Show that if a frictionless contact region is interposed between the bonded and separation regions, the singularity at the crack tip will be square-root singular in mode II, but not oscillatory. For simplicity, consider only the case where one body is rigid. Show also that there will then be a square-root singular contact pressure in the contact region, except when β = 0.

Chapter 36

Anisotropic Elasticity

Up to now, we have considered only isotropic materials, except in the development of certain general relations. However, many practical applications involve materials with significant anisotropy. This can arise by design, as in the development of fibrereinforced composites or metamaterials, or result from a manufacturing process such as rolling which causes grains to become elongated in the rolling direction. Also many naturally occurring materials are anisotropic, notably those resulting from a growth process or from biological adaption to the need for preferential strength or stiffness in a specific direction (for example the grain in wood). Crystalline materials are necessarily anisotropic at the scale of a single grain, though with random orientation of grains, a multigranular material will appear isotropic at length scales significantly larger than the average grain size. For this reason, it becomes more important to consider anisotropy when describing micromechanical effects — for example, in approximating the local state of stress due to a dislocation, or at a grain boundary.

36.1 The Constitutive Law The elasticity and compliance tensors (ci jkl , si jkl ) of equation (1.40) each have 81 components, but these are constrained by the symmetries (e.g. σi j = σ ji ) of the stress and strain tensors, and by Maxwell’s reciprocal theorem (see Chapter 38), so there are at most only 21 independent constants. However, this is still a sufficiently large number to impose a serious algebraic cost to the solution of elasticity problems for generally anisotropic materials. The algebraic relations can be simplified somewhat by using Voigt notation in which the six independent stresses are treated as components of a vector. The constitutive law can then be written in the form of the matrix equations σ=Ce;

e = Sσ ,

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 J. R. Barber, Elasticity, Solid Mechanics and Its Applications 172, https://doi.org/10.1007/978-3-031-15214-6_36

(36.1) 569

570

36 Anisotropic Elasticity

where σ = {σx x , σ yy , σzz , σ yz , σzx , σx y }T e = {ex x , e yy , ezz , 2e yz , 2ezx , 2ex y }T ,

(36.2) (36.3)

and C, S are 6 × 6 symmetric matrices derived from ci jkl , si jkl respectively. Notice the factor of 2 which is conventionally included in the shear strains in equation (36.3). Since the strain-energy density of equation (37.3) must be positive for all states of stress or strain, the matrices C and S are positive definite. Analytical methods of solution for anisotropic materials mostly depend on performing a linear transformation on the coördinates so as to reduce the governing equations to a standard form, usually the Laplace equation, or in two-dimensional problems, to allow solutions to be defined as holomorphic functions of a modified complex variable. We shall describe several illustrations of this technique in this chapter. If the characteristic equation defining these transformations has distinct roots (eigenvalues), a linear combination of the corresponding stress or displacement fields provides a general solution to the problem. In some cases, the equation has repeated roots and special techniques are required. A simple example where this occurs is discussed in §36.3.2 below.

36.2 Two-dimensional Solutions As in isotropic problems, we can identify states of deformation such that the displacements and hence the strains and stresses are independent of z. However, if all the coefficients in C, S are non-zero, the normal-shear coupling (e.g. between σzy and ex x ) implies that we should now expect non-zero antiplane shear stresses σzx , σzy and displacements u z even if the boundaries are loaded only by in-plane tractions. In effect, the plane strain problem of Chapters 4–13 and the antiplane problem of Chapter 15 become coupled. The most general case in which these problems remain uncoupled is that in which the z-plane is a plane of symmetry. It then follows that Ci j = 0 ; Si j = 0

if i ∈ {1, 2, 3, 6} and j ∈ {4, 5} or vice versa. (36.4)

Materials satisfying these conditions but with no additional symmetries are known as monoclinic. For the more general case where the z-plane is not a plane of symmetry, the in-plane and antiplane problems cannot be decoupled and we must satisfy all three equilibrium conditions (2.4), which in the absence of body forces have the reduced form

36.3 Orthotropic Material

∂σx y ∂σx x + =0 ; ∂x ∂y

571

∂σ yx ∂σ yy + =0 ; ∂x ∂y

∂σzy ∂σzx + =0. ∂x ∂y

(36.5)

Also, since the displacements are independent of z, we have 2ezx =

∂u z ∂u z ; 2ezy = ; ezz = 0 , ∂x ∂y

(36.6)

and since all z-derivatives in the compatibility equations (2.7) are zero, three of these are satisfied identically, and the remaining three reduce to ∂ 2 e yy ∂ 2 ex y ∂ 2 ex x + −2 =0 ; 2 ∂y ∂x∂ y ∂x 2

∂e yz ∂ezx − =0. ∂y ∂x

(36.7)

Broadly speaking, there are two approaches to the solution of the resulting twodimensional problem. Lekhnitskii1 satisfies the equilibrium equations by representing the stress components in terms of the Airy and Prandtl stress functions of equations (4.6, 18.5), and then develops a governing equation by substituting the resulting strains into the compatibility conditions (36.7). Stroh2 defines the displacement as a holomorphic function of a linearly transformed complex variable and obtains conditions for the transformation parameters from the equilibrium equations (36.5). We shall discuss both these methods in §§36.4, 36.5, below, but to introduce the topic we first consider the simpler uncoupled case of an orthotropic material under in-plane loading.

36.3 Orthotropic Material An orthotropic material is one with three mutually orthogonal symmetry planes. If we use these to define a Cartesian coördinate system, equation (36.1)2 takes the form ⎤ ⎡ S11 ex x ⎢ e yy ⎥ ⎢ S21 ⎢ ⎥ ⎢ ⎢ ezz ⎥ ⎢ S31 ⎢ ⎥ ⎢ ⎢ 2e yz ⎥ = ⎢ 0 ⎢ ⎥ ⎢ ⎣ 2ezx ⎦ ⎣ 0 2ex y 0 ⎡

S12 S22 S32 0 0 0

S13 S23 S33 0 0 0

0 0 0 S44 0 0

0 0 0 0 S55 0

⎤⎡ ⎤ σx x 0 ⎢ ⎥ 0 ⎥ ⎥ ⎢ σ yy ⎥ ⎢ σzz ⎥ 0 ⎥ ⎥⎢ ⎥ , ⎢ ⎥ 0 ⎥ ⎥ ⎢ σ yz ⎥ 0 ⎦ ⎣ σzx ⎦ σx y S66

(36.8)

and in view of the symmetry of the compliance matrix, there are nine independent elastic constants S11 , S12 , S13 , S22 , S23 , S33 , S44 , S55 , S66 . 1

S. G. Lekhnitskii, Theory of Elasticity of an Anisotropic Elastic Body, Holden-Day, San Francisco, 1963. 2 A. N. Stroh (1958), Dislocations and cracks in anisotropic elasticity, Philosophical Magazine, Vol. 3, pp. 625–646, A. N. Stroh (1962), Steady-state problems in anisotropic elasticity, Journal of Mathematics and Physics, Vol. 41, pp. 77–103.

572

36 Anisotropic Elasticity

Notice that this simplification of the constitutive law applies only when the coördinate system is chosen to align with the material symmetries. If we use equation (1.43) to rotate into another coördinate system, the coefficients Ci j , Si j will generally all be non-zero. This also implies that the directions of principal stresses and principal strains will not be aligned, except when these coincide with the material symmetries. Equation (36.8) and the condition ezz = 0 give σzz = −

S13 S23 σx x − σ yy , S33 S33

and hence the in-plane strains are ex x = β11 σx x + β12 σ yy ; e yy = β12 σx x + β22 σ yy ; 2ex y = β66 σx y , where, following Lekhnitskii, we define the 5 × 5 reduced compliance matrix β with components S3i S j3 ; i, j = 3 . (36.9) βi j = Si j − S33 Substituting these expressions into the compatibility condition (36.7)1 and using the Airy function representation σx x =

∂2φ ∂2φ ; σ = ; yy ∂ y2 ∂x 2

σx y = −

∂2φ ∂x∂ y

to satisfy the in-plane equilibrium equations (36.5)1,2 , we obtain the governing equation ∂4φ ∂4φ ∂4φ (36.10) β22 4 + (2β12 + β66 ) 2 2 + β11 4 = 0 . ∂x ∂x ∂ y ∂y This equation can be factorized as  β22

2 ∂2 2 ∂ + λ 1 ∂x 2 ∂ y2



2 ∂2 2 ∂ + λ 2 ∂x 2 ∂ y2

φ=0,

(36.11)

where λ21 , λ22 are the two roots of the quadratic equation β22 X 2 − (2β12 + β66 )X + β11 = 0 .

(36.12)

If the roots are both real and positive, we can then define two linear transformations of (x, y) ↔ (ξ, η) through x = ξ1 ; y = λ1 η1

and

x = ξ2 ; y = λ2 η2 ,

36.3 Orthotropic Material

573

so that equation (36.11) becomes  β22

∂2 ∂2 + ∂η1 2 ∂ξ1 2



∂2 ∂2 + ∂η2 2 ∂ξ2 2

φ=0,

and the general solution can be written as the sum of two harmonic functions in the respective transformed coördinates — i.e. φ(x, y) = φ1 (ξ1 , η1 ) + φ2 (ξ2 , η2 ) . Equivalently, we can write a general solution as

 φ =  f 1 (ζ1 ) + f 2 (ζ2 ) where

(36.13)

ζi = ξi + ıηi ,

and f 1 , f 2 are holomorphic functions.

36.3.1 Normal loading of the half-plane As an example, consider the problem where the half-plane y < 0 is loaded by purely normal tractions on the edge y = 0. The shear stress σx y = −

1 ∂ 2 φ1 ∂2φ 1 ∂ 2 φ2 =− − , ∂x∂ y λ1 ∂ξ1 ∂η1 λ2 ∂ξ2 ∂η2

and the line y = 0 corresponds to both η1 = 0 and η2 = 0, so the shear traction on this edge will be zero for all x if   λ1 λ1 y y − , f x, φ2 (ξ, η) = − φ1 (ξ, η) or φ = f x, λ2 λ1 λ2 λ2 where f is any harmonic function. The particular case of a concentrated force Fy at the origin, as in §12.3 and Figure 12.2, can be solved using the harmonic function   

2 η 2 , f (ξ, η) = C η ln ξ + η + 2ξ arctan ξ where C is an arbitrary constant which can be related to Fy by considering the equilibrium of the strip −h < y < 0. We obtain C=

λ1 Fy , 2π(λ2 − λ1 )

and the corresponding stress components are

(36.14)

574

36 Anisotropic Elasticity

σx x =

Fy (λ1 + λ2 )x 2 y Fy (λ1 + λ2 )y 3 ; σ = yy 2 2 π(y 2 + λ1 x 2 )(y 2 + λ2 x 2 ) π(y 2 + λ21 x 2 )(y 2 + λ22 x 2 ) σx y =

Fy (λ1 + λ2 )x y 2 . π(y 2 + λ21 x 2 )(y 2 + λ22 x 2 )

(36.15)

36.3.2 Degenerate cases This solution technique cannot be used if the two roots of equation (36.12) are equal (λ21 = λ22 ), since the two functions in equation (36.13) could then be combined to form a single arbitrary harmonic function, and this does not provide enough degrees of freedom (e.g. arbitrary constants) to satisfy the boundary conditions. However in this case, equation (36.11) can be converted to the biharmonic equation by a single transformation y = λη and the solution can then be obtained as in Chapters 4–13, a special case being that of an isotropic material, for which λ21 = λ22 = 1. It is interesting to note that if we set λ21 = λ22 = 1 in the final expressions for the stresses (36.15), we recover the Cartesian equivalent of the isotropic Flamant solution. In other words, if we pretend that the material is orthotropic with λ21 = λ22 , and proceed to the isotropic limit only at the last step, the correct solution is obtained, even though intermediate steps such as equation (36.14) are ill-defined in this limit. This situation is clearly similar to that described in §§8.3.3, 10.3.1, where an otherwise general technique breaks down at one or more special values of a parameter. We could formalize the limiting process in the present case by defining (say) λ2 = λ1 +  and taking the limit as  → 0. It is significant that the constant C in (36.14) is unbounded when λ22 → λ21 which compensates for the harmonic functions becoming equal, exactly as in equation (8.30). An alternative approach is to obtain the limit by differentiation, defining a second function by φ2 =

∂ η ∂ f 2 (ξ, η)) = − f (ξ, η) . ∂λ λ ∂η

Since the spatial derivative of a harmonic function is itself harmonic, this is equivalent to replacing (36.13) by φ(x, y) = φ1 (ξ, η) + ηφ2 (ξ, η) , which defines a general two-dimensionally biharmonic function in (ξ, η) coördinates if φ1 , φ2 are arbitrary harmonic functions. Related degeneracies occur for more general anisotropic materials in both Lekhnitskii and Stroh solutions when two or more of the transformations required for the solution coincide. They can be resolved by similar techniques to those described above.

36.4 Lekhnitskii’s Formalism

575

36.4 Lekhnitskii’s Formalism We now consider the case of a general anisotropic material, where any or all of the coefficients in C, S may be non-zero. We define the stress components σx x =

∂2φ ∂2φ ∂2φ ; σ = ; σ = − yy x y ∂ y2 ∂x 2 ∂x∂ y ∂ψ ∂ψ ; σzy = − , σzx = ∂y ∂x

(36.16)

where φ, ψ are the Airy and Prandtl stress functions of Chapters 4, 17 respectively3 . If there is no body force, this representation satisfies the equilibrium equations identically, so it remains to satisfy the non-trivial compatibility equations (36.7). As in §36.3, we use the condition ezz = 0 to eliminate the stress component σzz , so that there are only five independent stress and strain components in equations (36.2, 36.3), which we combine as ς = {σx x , σ yy , σ yz , σzx , σx y }T ;  = {ex x , e yy , 2e yz , 2ezx , 2ex y }T . The constitutive law (36.1)2 then reduces to  = βς ,

(36.17)

where the reduced compliance matrix β is defined by equation (36.9). Substituting (36.17) into the compatibility equations (36.7) and using the stress function representation (36.16) for the stress components, we then obtain L4 φ + L3 ψ = 0 ; L3 φ + L2 ψ = 0 ,

(36.18)

where the differential operators ∂2 ∂2 ∂2 + β55 2 − 2β45 2 ∂x ∂x∂ y ∂y 3 3 ∂ ∂3 ∂ ∂3 − (β56 + β14 ) L3 ≡ −β24 3 + (β25 + β46 ) 2 + β 15 ∂x ∂x ∂ y ∂x∂ y 2 ∂ y3 4 4 4 4 ∂ ∂ ∂ ∂4 ∂ + (2β12 + β66 ) 2 2 − 2β16 L4 ≡ β22 4 − 2β26 3 + β . 11 ∂x ∂x ∂ y ∂x ∂ y ∂x∂ y 3 ∂ y4

L2 ≡ β44

If the stiffness matrix satisfies the monoclinic condition (36.4), the differential operator L3 ≡ 0, and the problem then decouples into an in-plane problem and an antiplane problem governed by L4 φ = 0 and L2 ψ = 0 respectively. In particular, for the orthotropic case, the in-plane equation L4 φ = 0 reduces to (36.10) Notice that we are here using ψ for the Prandtl function, to avoid possible confusion between φ and ϕ when writing solutions by hand.

3

576

36 Anisotropic Elasticity

For the coupled case, simultaneous solution of equations (36.18) shows that both φ and ψ must satisfy the sixth order partial differential equation L6 {φ, ψ} = 0

where

L6 ≡ L4 L2 − L3 L3 .

(36.19)

36.4.1 Polynomial solutions Polynomial solutions to problems in rectangular coördinates can be found using exactly the same techniques as in Chapter 5. We first identify the order of the polynomial in φ as in §5.2.2, noting that for general anisotropy we cannot appeal to geometric symmetry to eliminate any of the terms. Equation (36.18)2 then implies that we shall also generally require a polynomial for ψ of one order less than that for φ.

Example As a simple example, we consider the plane strain equivalent of Figure 5.2, where the surfaces y = ±b are traction free, and the end x = 0 is loaded by a shear force F per unit length in the negative y-direction. Since we anticipate non-zero contributions in all the stress components, the weak boundary conditions on x = 0 must include both forces and moments, comprising4 

b −b

 σx x (x, 0) dy = 0 ;



b

σx y (x, 0) dy = F ;

−b



b −b

σx x (x, 0) y dy = 0 .

b −b

σx z (x, 0) dy = 0 ;

(36.20)

We anticipate a dominant stress component σx x ∼ x y, so we start with a fourth degree polynomial for φ and hence a third degree polynomial for ψ — i.e.

4

It is tempting here to add the weak condition 

b −b

σx z (x, 0) y dy = 0 ,

corresponding to the absence of a distributed torque on the end x = 0. However, if such a torque exists, its release would imply twist of the cross section, which violates the plane strain condition ∂u y /∂z = 0. For more discussion of the torsion problem, see §36.6.2 below.

36.4 Lekhnitskii’s Formalism

577

φ = A1 x 4 +A2 x 3 y + A3 x 2 y 2 + A4 x y 3 + A5 y 4 + A6 x 3 + A7 x 2 y + A8 x y 2 + A9 y 3 + A10 x 2 + A11 x y + A12 y 2 , ψ = B1 x 3+B2 x 2 y + B3 x y 2 + B4 y 3 + B5 x 2 + B6 x y + B7 y 2 + B8 x + B9 y . (36.21) The stress components are then defined by equations (36.16), and substitution into the boundary conditions σ yx = σ yy = σ yz = 0 ; all x, y = ±b , yields six polynomial equations. Two additional equations are obtained by substituting (36.21) into equations (36.18). Equating coefficients of powers of x and y in these equations, and imposing the end conditions (36.20), we obtain a total of 29 linear algebraic equations which can be solved for the 21 constants Ai , Bi . It will be appreciated that even a problem as simple as this is practical only when using software for the algebraic calculations! The final stress field is given by5   3y 2 y2 3F x y Fα1 3F ; σx y = ; σ yy = 0 1− 2 1− 2 σx x = + 2b3 b b 4b b   Fα3 3Fα2 3Fα2 x y 3y 2 y2 ; σ yz = − σx z = 1− 2 − 1− 2 , b b 2b3 4b b where 2 2 − β16 β55 β55 β15 β56 + β55 β14 β15 − β45 β15 β15 ; α2 = 2 β55 2β55 (β55 β11 − β15 ) 2 ) 2β15 (β55 β16 + β45 β11 ) − (β56 + β14 )(β55 β11 + β15 α3 = 2 4β55 (β55 β11 − β15 )

α1 =

Notice that the anisotropy introduces additional terms into these stress components compared with equations (5.22), and in particular that the resulting solution has no symmetry about y = 0.

36.4.2 Solutions in linearly transformed space It can be verified by substitution that equation (36.19) has a particular solution of the form where ζ = x + py , φ = f (ζ) + f (ζ) = 2  {( f (ζ)}

The final stress component σzz can be found by imposing the condition ezz = 0, but this step is omitted here in the interests of brevity.

5

578

36 Anisotropic Elasticity

if p is chosen to satisfy the sixth-order polynomial equation 6 ( p) ≡ 4 ( p)2 ( p) − 3 ( p)3 ( p) = 0 ,

(36.22)

and 2 ( p) = β44 − 2β45 p + β55 p 2 3 ( p) = −β24 + (β25 + β46 ) p − (β56 + β14 ) p 2 + β15 p 3 4 ( p) = β22 − 2β26 p + (2β12 + β66 ) p 2 − 2β16 p 3 + β11 p 4 . Lekhnitskii6 has shown that the polynomial equation (36.22) has no real roots, and since the coefficients are all real, the six solutions must comprise three complex conjugate pairs, each of which defines a linear transformation of the in-plane coördinates. For example, if p = b + ıc and we write ζ = ξ + ıη, then equation (36.23)2 maps (x, y) into (ξ, η) through ξ = x + by ; η = cy . In the following derivations, we shall sometimes omit the real part operator {·} in the interests of brevity. Suppose we choose one of the roots pα of equation (36.22) and write φα = 2  { f (ζα )} ;

ψα = 2  {g(ζα )} ,

where ζα = x + pα y. Then (36.18)2 requires that L2 (ψα ) = 2 ( pα )g  (ζα ) = −L3 (φα ) = −3 ( pα ) f  (ζα ) and a particular solution is g(ζα ) = −λα f  (ζα )

where

λα =

4 ( pα ) 3 ( pα ) = . 2 ( pα ) 3 ( pα )

Using these results, we can then write a general solution as  3   3     φ = 2 f α (ζα ) ; ψ = −2  λα f α (ζα ) . α=1

α=1

The three independent holomorphic functions f α (ζα ), α = (1, 2, 3) then allow us to satisfy three boundary conditions at each point on the boundary of the twodimensional domain. Notice that the equation ζα = x + pα y defines a different linear transformation of the two-dimensional space for each value of α. In particular, the circle transforms into three different ellipses, and hence requires the use of the conformal mapping of 6

S. G. Lekhnitskii, loc. cit.

36.5 Stroh’s Formalism

579

§§19.7, 20.7.1. However, straight lines remain straight under linear transformation, so (for example) boundary-value problems for wedge-shaped domains can be solved using techniques similar to those in Chapter 11.

36.5 Stroh’s Formalism Stroh’s formalism7 starts by considering the conditions under which the equilibrium conditions can be satisfied by a displacement field of the form8 u = a f (ζ) + a f (ζ) = 2  {a f (ζ)}

where

ζ = x + py .

(36.23)

The displacement gradients are then given by

 ∂u = 2  a f  (ζ) ; ∂x

 ∂u = 2  a p f  (ζ) ; ∂y

∂u =0, ∂z

and the strain components are





 ex x = 2  a1 f  (ζ) ; e yy = 2  a2 p f  (ζ) ; 2ex y = 2  (a1 p + a2 ) f  (ζ)

 2ezx = 2  a3 f  (ζ) ;

 2e yz = 2  a3 p f  (ζ) ; ezz = 0 .

(36.24)

Substituting these expressions into the constitutive law (36.1)1 and reverting to the index notation for stress components, we obtain   

 Q ik + p Rik ak f  (ζ) ; σi2 = 2  Rki + pTik ak f  (ζ) , (36.25) where i, k = 1, 2, 3 and ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ C11 C16 C15 C16 C12 C14 C66 C26 C46 Q = ⎣ C61 C66 C65 ⎦ ; R = ⎣ C66 C62 C64 ⎦ ; T = ⎣ C62 C22 C42 ⎦ , C51 C56 C55 C56 C52 C54 C64 C24 C44 σi1 = 2 

7



This method is based on earlier results due to J. D. Eshelby, W. T. Read and W. Shockley (1953), Anisotropic elasticity with applications to dislocation theory, Acta Metallurgica, Vol. 1(3), pp. 251– 259, but Stroh’s contributions to its development are so significant that it is now generally referred to by his name. For a much more detailed and rigorous development of the method, see T. C. T. Ting, Anisotropic Elasticity: Theory and Applications, Oxford University Press, New York, 1996. 8 Notice that although this representation has some similarity to that used in Chapter 20, we cannot take the further step of identifying the displacement components with the real and imaginary parts of a complex function because there are now three displacement components u x , u y , u z .

580

36 Anisotropic Elasticity

or equivalently, Q ik = ci1k1 , Rik = ci1k2 , Tik = ci2k2 . We can also obtain an expression for the stress component σ33 from (36.24) and (36.1)1 , but this plays no rôle in the solution procedure. The three equilibrium equations (36.5) can be written concisely in the form ∂σi1 ∂σi2 + =0 ∂x1 ∂x2

∂σi1 ∂σi2 +p =0, ∂ζ ∂ζ

and hence

and these will be satisfied by the stress components (36.25) if   

Q + p R + R T + p 2 T a f  (ζ) = 0 .

(36.26)

(36.27)

Equation (36.27) has non-trivial solutions if and only if the determinant  

  Q + p R + R T + p2 T  = 0 , (36.28) and this yields a sextic equation for p which for physically realistic material properties always has three pairs of complex conjugate roots, those with positive imaginary part being denoted by p1 , p2 , p3 respectively. Barnett and Kirchner9 have shown that the roots of equation (36.28) are identical to those of Lekhnitskii’s polynomial equation (36.22). If the roots (eigenvalues) p1 , p2 , p3 are distinct, a general solution of the twodimensional problem can then be constructed by superposition in the form  u = 2

3 

 a(α) f α (ζα )

,

(36.29)

α=1

where ζα = x + pα y and a(α) is the eigenvector of (36.27) corresponding to the eigenvalue pα . The three independent holomorphic functions f α provide sufficient generality to satisfy three conditions at each point on the boundary and hence can represent any two-dimensional stress state. The equilibrium equation (36.26)1 implies that the stress components can be expressed in terms of a vector stress function φ, where σi1 = −

∂φi ∂φi ; σi2 = ; i = 1, 2, 3 . ∂x2 ∂x1

(36.30)

Equations (36.25) then imply that  φ = 2

3 

 (α)

b

f α (ζα )

,

(36.31)

α=1

9

D. M. Barnett and H. O. K. Kirchner (1997), A proof of the equivalence of the Stroh and Lekhnitskii sextic equations for plane anisotropic elastostatics, Philosophical Magazine A, Vol. 76 (1), pp. 231–239.

36.5 Stroh’s Formalism

where

581

 1 b(α) = R T + pα T a(α) = − ( Q + pα R) a(α) . pα

(36.32)

Since a(α) , b(α) each comprise sets of three vectors (α = 1, 2, 3), it is often convenient to combine them into matrices by defining ⎤ a1(1) a1(2) a1(3) A = ⎣ a2(1) a2(2) a2(3) ⎦ ; a3(1) a3(2) a3(3) ⎡

⎡

⎤ b1(1) b1(2) b1(3) B = ⎣ b2(1) b2(2) b2(3) ⎦ . b3(1) b3(2) b3(3)

(36.33)

Equations (36.29, 36.31) can then be written concisely as u = 2  { A f } ; φ = 2  {B f }

f = { f 1 (ζ1 ), f 2 (ζ2 ), f 3 (ζ3 )}T . (36.34) In many applications, the form of the three functions f α (ζα ) will be the same, so that   f = f (ζ) q

where



where

   f (ζ) = diag f (ζ1 ), f (ζ2 ), f (ζ3 ) ,

(36.35)

and q is a constant (position-independent) vector.

36.5.1 The eigenvalue problem Routine matrix operations on equations (36.32) show that the vectors a (α) , b(α) satisfy the eigenvalue equations

N1 N2 N3 N4

   a a =p b b

and

[ b, a ]

N1 N2 N3 N4

 = p [ b, a ] ,

(36.36)

where N 1 = −T −1 R T ; N 2 = T −1 ; N 3 = RT −1 R T − Q ; N 4 = N 1T . In other words, [a(α) , b(α) ]T and [b(α) , a(α) ] are respectively the right and left eigenvectors of the matrix  N1 N2 . N= N3 N4 It follows that the eigenfunctions corresponding to distinct eigenvalues α = i and α = j with i = j satisfy the orthogonality conditions

582

36 Anisotropic Elasticity



    a( j)   (i)  (i) (i)  a( j)  a( j) (i) = b =0. b ,a , a b , a = ( j) b( j) b( j) b (i)

(i)

In other words, the matrix product 

B T AT T T B A



A A B B



is diagonal. Since equations (36.36) are homogeneous, any linear (possibly complex) multiplier of the eigenfunctions will also satisfy them, provided pα satisfies (36.28), so it is convenient to normalize the matrices A, B so as to satisfy the condition10 

B T AT T T B A



A A B B



 =

I 0 0 I

 ,

(36.37)

where I is the 3 × 3 identity matrix δαβ . It then follows that the two 6 × 6 matrices on the left-hand side of (36.37) are mutual inverses, and also therefore that the matrix product commutes.

36.5.2 Solution of boundary-value problems The matrices Q, R, T , and hence the eigenvalues pα and vectors a(α) , b(α) depend only on the Voigt stiffness matrix C, so they can be calculated once and for all for a given anisotropic material. The problem is therefore reduced to a boundary-value problem similar to those treated in Chapter 20. However, as with the Lekhnitskii formulation of §36.4.2, three different conformal mappings will generally be required11 , one for each eigenvalue pα . An important exception comprises problems for the infinite plane or the half-plane, since the boundary x2 = 0 (y = 0) corresponds to the x-axis for all values of p in equation (36.23)2 . This class includes two-dimensional contact problems (Chapter 12), the Kelvin problem (§13.1), dislocations (§13.2) and plane crack problems (§13.3). Suppose the tractions ti (x) are known functions of x on the boundary y = 0 of the half-plane y > 0. Equations (36.30)2 and (36.31) then show that  ti (x) = −2 

3 

 bi(α) f α

(x)

,

(36.38)

α=1

However, notice that with this normalization, A and B will have the dimensions of stress−1/2 and stress1/2 respectively. 11 Problems of this kind are discussed by T. C. T. Ting, loc. cit., Chapter 10. 10

36.5 Stroh’s Formalism

583

where the negative sign arises because the unit outward normal to the surface is defined by −δi2 . Equation (36.38) comprises three simultaneous equations for the  (19.39) to determine the holomorphic three functions

f α(x). We can then use

equation   functions f α ζ α , after which f α ζ α is determined by integration. The complete stress and displacement fields are then given by equations (36.29–36.31).

Example Suppose the half-plane y > 0 is loaded by a uniform traction si in −c < x < c, so that ti (x) = si g(x), where g(x) = 1 ; −c < x < c = 0 ; |x| > c . Equation (36.38) will then be satisfied if f α (ζα ) = −qα h(ζα )

where

Bq = s

¯ h(x) + h(x) = g(x) .

and

Using the results from Example 1 of §19.6.3, we then obtain f α (ζ α ) = −

 ζ −c qα ln α 2πı ζα + c

and hence, on integration, f α (ζ α ) =

 qα  (ζ α + c) ln(ζ α + c) − (ζ α − c) ln(ζ α − c) . 2πı

36.5.3 The line force solution In the limit where s = F/2c and c → 0 in the above example, we recover the solution for a line force F at the origin, which is f α (ζ α ) =

qα ln(ζα ) 2πı

where

Bq = F .

The corresponding displacement u and the stress function φ are then given by equations (36.34, 36.35) as u=

    1  1 

A ln(ζ) q ; φ = B ln(ζ) q . π π

In particular, the surface displacement is

584

36 Anisotropic Elasticity

 1

1

{ Aq ln(x)} = AB −1 F ln(x) π π 1 = −  { D F ln(x)} , π

u(x, 0) =

where we define D ≡ M −1 = ı AB −1 as the inverse of the impedance tensor 12 M = −ı B A−1 . Since ln(x) = ln(|x|) + ıπ H (−x) and F is real, we then have 1 1 u(x, 0) = −  { D} F ln(|x|) − { D}F sgn(x) , π 2

(36.39)

where we have added a rigid-body translation to restore symmetry, as in Chapter 12. Maxwell’s reciprocal theorem §38.1 shows that  { D} must be symmetric and { D} must be skew symmetric, so the matrices D and M are Hermitian13 (i.e. M T = M).

36.5.4 Internal forces and dislocations A similar approach can be used for the Kelvin problem of §13.1. If a concentrated force F per unit length (along the x3 -axis) acts at the origin in the infinite space, we know from equilibrium considerations and self-similarity that the stress components will decay with r −1 , which in view of the representation (36.30, 36.31) implies that the functions f α (ζ α ) should be multiples of the holomorphic function ln(ζ α ). Using (36.34, 36.35), we therefore write u = 2  { A ln(ζ) q} ; φ = 2  {B ln(ζ) q} .

(36.40)

Equilibrium of a cylindrical region of radius a then gives the condition  Fi +

2π

σi j n j adθ = 0

where

n = {cos θ, sin θ, 0}

(36.41)

0

is the outward normal. Also, since points on the circle are defined by x1 = a cos θ, x2 = a sin θ, we have ∂x1 = −a sin θ ; ∂θ

∂x2 = a cos θ . ∂θ

Using these results and (36.30), we can write (36.41) in the form  Fi + 0 12 13

2π

 ∂φi ∂x2 ∂φi ∂x1 − − dθ = 0 ∂x2 ∂θ ∂x1 ∂θ

or

2π Fi = φi 0 .

See T. C. T. Ting loc. cit., Chapter 5. This can be proved directly from the properties of the matrices A, B.

(36.42)

36.5 Stroh’s Formalism

585

From the properties of ln(ζ) and equations (36.40, 36.42), we conclude that F = 2  {2πı Bq} = 2 {Bh}

where

h = 2πı q .

(36.43)

We also note that the displacement defined by equation (36.29) will generally exhibit a discontinuity on the cut θ = 0, 2π, defining a dislocation with Burgers’ vector14 2π B ≡ {Bx , B y , Bz }T = − u0 = −2 {2πı Aq} = −2 { Ah} .

(36.44)

Equations (36.43, 36.44) can be combined in the matrix equation

A A B B

   h −B = , F h

and if the force F and the Burgers’ vector B are prescribed, this equation can be solved, using the inverse property (36.37), to obtain      h −B B T AT . = T T F h B A

36.5.5 Planar crack problems Equations (36.40, 36.43, 36.44) can be used exactly as in §13.3.2 to solve the problem of the plane crack −a < x < a, y = 0 in a body loaded by uniform tractions σi2 → T at the remote boundaries. Alternatively, the corrective solution — i.e. the solution for the same crack loaded only by tractions −T , T on the crack faces y = 0− , y = 0+ respectively — is easily obtained in the form u = 2  { A  f (ζ) q} ; φ = 2  {B  f (ζ) q} , with f (ζ) =



ζ 2 − a2 − ζ .

The tractions on the plane y = 0− are then given by t(x, 0− ) = σi2 (x, 0) = 14

   |x| ∂φ (x, 0) = 2  Bq √ −1 , ∂x x 2 − a2

not to be confused with the matrix B of equation (36.33). Ting loc. cit. defines the Burger’s vector as a displacement discontinuity on the negative x-axis — i.e. as u|π−π —, leading to the opposite sign on B. Here we follow the same convention as in Chapter 13, where the discontinuity is placed on the positive x-axis.

586

36 Anisotropic Elasticity

and this satisfies the boundary condition t(x, 0− ) = −T in −a < x < a if Bq is real and given by B −1 T . −2Bq = −T or q = 2 The corrective tractions on the plane y = 0 outside the crack (|x| > a) are then given by   |x| |x| t(x, 0) = 2Bq √ −1 = T √ −1 . x 2 − a2 x 2 − a2 Notice that this traction and hence also the stress-intensity factors depend only on the far-field loading and are independent of the anisotropic constants.

36.5.6 The Barnett-Lothe tensors The method of solution described in this section requires the determination of the complex eigenvalues pα and eigenvectors a(α) , b(α) from equations (36.28, 36.27, 36.32). However, in many problems, important results can be expressed in terms of the real Barnett-Lothe tensors 

S = ı 2 AB T − I ;

H = 2ı A AT ;

L = −2ı B B T ,

where A, B are normalized so as to satisfy equation (36.37). Furthermore, S, H and L can be determined from the real matrices Q, R and T without explicit solution of the eigenvalue problem. We obtain15 S=−

1 π



 1 π −1 T −1 (θ)R T (θ)dθ ; H = T (θ)dθ π 0 0   1 π L=− R(θ)T −1 (θ)R T (θ) − Q(θ) dθ , π 0 π

where 

Q(θ) = Q cos2 θ + R + R T sin θ cos θ + T sin2 θ R(θ) = R cos2 θ + (T − Q) sin θ cos θ − R T sin2 θ 

T (θ) = T cos2 θ − R + R T sin θ cos θ + Q sin2 θ .

15

For a proof of this result, see T. C. T. Ting loc. cit., Chapter 7, or A. F. Bower, Applied Mechanics of Solids, CRC Press, Boca Raton, 2010, §5.5.11.

36.6 End Loading of the Prismatic Bar

587

Once S, H and L have been determined, the impedance tensor and its inverse can be obtained as M = −ı B A−1 = H −1 + ı H −1 S M −1 = D = ı AB −1 = L −1 − ı SL −1 .

(36.45)

These results enable us to write equation (36.39) (for example) in terms of real matrices as 1 1 u(x, 0) = − L −1 F ln(|x|) + SL −1 F sgn(x) . π 2

36.6 End Loading of the Prismatic Bar If a long bar of uniform cross section is loaded only at the ends by a torque, a bending moment or an axial force, the stresses distant from the ends will be independent of z, but the displacements will contain terms that vary with z. Thus, these problems do not fall into the category defined in §36.2 and treated in §§36.3–36.5.

36.6.1 Bending and axial force If we postulate an elementary stress state where the only non-zero stress component is σzz = Ax + By + C, the resulting strains will be linear functions of x, y and hence all six compatibility equations will be satisfied identically. Also, the tractions on the lateral surfaces of the bar will be zero, so this elementary stress state defines an exact solution to the problem of a prismatic bar loaded only by equal and opposite bending moments and axial forces at the two ends16 . In particular, the stress components depend only on the force and moment resultants and the cross section of the bar and are independent of the elastic constants. However, if the monoclinic condition (36.4) is not satisfied — i.e. if the in-plane and antiplane problems are coupled — the axial displacement u z will generally be a quadratic function of x, y and hence all initially plane sections will warp into identical quadratic surfaces. Also the bending and torsion problems are then coupled — bending moments cause twist and torques cause the axis of the bar to bend.

16

If the tractions on the ends differ from those implied by this solution but have the same force and moment resultants, we anticipate exponentially decaying modifications to the stress fields near the ends that can be found by the technique introduced in Chapter 6.

588

36 Anisotropic Elasticity

36.6.2 Torsion Suppose the bar transmits a torque T and the axis is constrained to remain straight by the imposition of appropriate bending moments on the ends. We assume that there will be a uniform twist per unit length β, and, as in equation (17.1), this implies additional z-dependent terms −βzy and βzx in u x , u y respectively. Equation (36.6) is therefore modified to 2ezx =

∂u z ∂u z − β y ; 2ezy = + βx , ∂x ∂y

and the second compatibility equation (36.7)2 becomes ∂e yz ∂ezx − +β =0. ∂y ∂x

(36.46)

Substitution of (36.17) into the compatibility equations (36.7)1 and (36.46) then yields the differential equations L4 φ + L3 ψ = 0 ; L3 φ + L2 ψ = −2β . As in Chapter 17, a convenient method of solution is to write ψ = ψ H + ψ P , where the particular solution ψ P satisfies the equation L2 ψ P = −2β . Any convenient quadratic function of x, y can be chosen for ψ P , in which case L3 ψ P = 0. The functions φ and ψ H will then satisfy (36.18), and general solutions can be obtained as in §36.4.2. It is arguably simpler conceptually to (i) determine the stress field due to ψ P alone (with φ = 0), (ii) find the tractions implied on the lateral surfaces, and then (iii) use these to define a torsion-free corrective solution that can be treated using the method of §36.4. If the in-plane and antiplane problems are coupled (L3 = 0), the stress component σzz (x, y) will be non-zero, so non-zero force and moment resultants F, Mx , M y are generally required on the ends of the bar. If the ends are actually traction free, an approximate solution can be obtained by cancelling these resultants by superposing an elementary stress state, as discussed in §3.1.1.

36.7 Three-dimensional Problems Three-dimensional problems for anisotropic bodies are extremely challenging and very little progress can be made using purely analytical methods. The potential func-

36.7 Three-dimensional Problems

589

tion approach used in Chapters 21–33 has no anisotropic counterpart except where the anisotropy is axisymmetric, including the special case of transverse isotropy (see §36.8). Lekhnitskii’s formalism can be used for the problem of an end loaded cantilever17 — i.e. for the anisotropic solution of the problems considered in Chapter 18 — and the recursive technique of Chapter 30 can be applied to Stroh’s formalism to provide the general solution of the prismatic bar loaded by polynomial tractions on the lateral surfaces18 .

36.7.1 Concentrated force on a half-space If a concentrated normal force is applied at the origin on the otherwise traction-free surface of the half-space z > 0, the stress state is self-similar and equilibrium considerations show that the stress components will vary with R −2 and the displacement components with R −1 . In particular, the normal surface displacements must take the form Fh(θ) u z (r, θ, 0) = , (36.47) r where h(θ) depends on the elastic constants ci jkl or equivalently on the stiffness matrix C of equation (36.1). Willis19 used this argument to show that the contact area in Hertzian contact remains elliptical for general anisotropy, and the contact pressure distribution is still given by equation (34.4). This result is easily verified using the field-point integration method of §34.1.1 since the function h(θ) appears only in the outer integral in the anisotropic counterpart of equation (34.7). However, the parameters a, b, p0 defining the semi-axes of the ellipse and the maximum contact pressure p0 depend on the function h(θ) in equation (36.47). Willis solved the concentrated force problem using a double Fourier transform, but an arguable more elegant solution was provided by Sveklo20 , using the SmirnovSobolev transform. A remarkable result from this solution is that h

17

D22 π! = 2 2π

where

D = M −1 = L −1 − ı SL −1

(36.48)

S. G. Lekhnitskii, loc. cit., Chapter 5. J. R. Barber and T. C. T. Ting (2007), Three-dimensional solutions for general anisotropy, Journal of the Mechanics and Physics of Solids, Vol. 55, pp. 1993–2006. 19 J. R. Willis (1966), Hertzian contact of anisotropic bodies, Journal of the Mechanics and Physics of Solids, Vol. 14 (3), pp. 163–176. 20 V. A. Sveklo (1964), Boussinesq type problems for the anisotropic half-space, Journal of Applied Mathematics and Mechanics, Vol. 28 (5), pp. 1099–1105. 18

590

36 Anisotropic Elasticity

is the inverse of the impedance matrix21 . In other words, the displacement at θ = π/2 is proportional to the multiplier on the logarithmic term in the corresponding two-dimensional solution of equation (36.39) with normal loading only. Thus an alternative expression for h(θ) is π! 1 D22 θ − , h(θ) = 2π 2 where D(α) is the inverse impedance matrix in a coördinate system rotated through α about the z-axis.

36.8 Transverse Isotropy For a transversely isotropic material, the most general form of Hooke’s law can be written ⎡ ⎤⎡ ⎤ ⎡ ⎤ ex x C11 C12 C13 0 0 σx x 0 ⎢ σ yy ⎥ ⎢ C21 C11 C13 0 0 ⎥ ⎢ e yy ⎥ 0 ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ σzz ⎥ ⎢ C31 C31 C33 0 0 ⎥ ⎢ ezz ⎥ 0 ⎢ ⎥⎢ ⎥=⎢ ⎥ (36.49) ⎢ σ yz ⎥ ⎢ 0 0 0 C44 0 ⎥ ⎢ 2e yz ⎥ . 0 ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎣ σzx ⎦ ⎣ 0 0 0 0 C44 ⎦ ⎣ 2ezx ⎦ 0 σx y 2ex y 0 0 0 0 0 (C11 − C12 )/2 This suggests that solutions might be obtained using an appropriate linear scaling of the z-axis, so in the spirit of §21.1, we look for conditions under which a solution can be defined in terms of a displacement function φ through the relations ux =

∂φ ∂φ ∂φ ; uy = ; uz = λ , ∂x ∂y ∂z

where λ is a scalar parameter. Substituting these definitions into the strain-displacement relations and the resulting strains into (36.49), we obtain σx x = C11

∂2φ ∂2φ ∂2φ ∂2φ ∂2φ ∂2φ + C + C λ ; σ = C + C + C λ 12 13 yy 12 11 13 ∂x 2 ∂ y2 ∂z 2 ∂x 2 ∂ y2 ∂z 2 ∂2φ ∂2φ ∂2φ σzz = C13 2 + C13 2 + C33 λ 2 ∂x ∂y ∂z

σ yz = C44 (1 + λ)

∂2φ ∂2φ ∂2φ ; σzx = C44 (1 + λ) ; σx y = (C11 − C12 ) . ∂ y∂z ∂z∂x ∂x∂ y

These stress components will satisfy the three equilibrium equations (2.4) with no body force if φ satisfies the two equations 21

See §36.5.3 and equation(36.45). Also, note that D is Hermitian, so D22 is real.

36.8 Transverse Isotropy

591

  ∂2φ C11 ∇12 φ + C13 λ + C44 (1 + λ) =0 ∂z 2   ∂2φ C13 + C44 (1 + λ) ∇12 φ + C33 λ 2 = 0 , ∂z 

where we write ∇12 ≡

∂2 ∂2 + 2 2 ∂x ∂y

(36.50) (36.51)



for the two-dimensional Laplacian operator. The two conditions (36.50, 36.51) become identical if we choose λ such that    C13 + C44 (1 + λ) C13 λ + C44 (1 + λ) = C11 C33 λ . This is a quadratic equation for λ with two roots λ1 , λ2 = c ±



c2 − 1

where

c=

2 C11 C33 − C13 −1. 2C44 (C13 + C44 )

When λ = λi (i = 1, 2), the two equations (36.50, 36.51) both reduce to ∇12 φ + pi where pi =

∂2φ =0, ∂z 2

(36.52)

C44 + (C13 + C44 )λi . C11

If we then make the further transformation √ z = Z i pi , equation (36.52) is reduced to Laplace’s equation ∂2φ ∂2φ ∂2φ + + =0 2 2 ∂x ∂y ∂ Zi 2 in the coördinate system (x, y, Z i ). A more general solution can be constructed by superposing the solutions associated with functions φ1 , φ2 , corresponding to the eigenvalues λ1 , λ2 . These two functions can then be used in the same way as Solutions A and B of Table 22.1 to formulate boundary-value problems for the half-space. They can also be conveniently combined to construct solutions analogous to Solutions F and G of Table 22.3 in which either the shear traction or the normal traction is identically zero throughout the surface z = 0. For non-axisymmetric problems, a third potential function is required, analogous to Solution E, of Table 22.1. The development of this solution is left as an exercise for the reader (Problem 36.11).

592

36 Anisotropic Elasticity

In the special case of isotropy, c = 1, so λ1 = λ2 = 1 and p1 = p2 = 1. However, as in §36.3.2, if we approach the isotropic case as a limit c → 1+ , two independent solutions in terms of harmonic functions can be obtained.

Problems 36.1. Use the Prandtl stress function ϕ of equations (17.4) to satisfy the antiplane equilibrium equation (36.5)3 and then use the compatibility equation (36.7)2 to develop the governing equation for ϕ for the orthotropic material of equation (36.8). Use this formulation to find the stress components for the problem of Figure 15.2. 36.2. Use the transformation equations (1.43) and the two dimensional direction   cosines (1.14) to find the components C16 , C26 for the orthotropic material of equation (36.8) in a coördinate system rotated through an angle θ about the z-axis. Hence show that these coefficients are zero for all θ if and only if C11 = C22

and

2C66 = C11 − C12 .

If these condition are not satisfied, how many planes (values of θ in 0 ≤ θ < π) exist   = C26 = 0? on which C16 36.3. Transform the stress components (36.15) into polar coördinates (r, θ) centred on the point of application of the force Fy and hence show that σθr = σθθ = 0 for all (r, θ) as in the isotropic case. Can you find a proof of this without using the orthotropic solution of §36.3.1? Use your results to make plots of the function σrr (r, θ)/σrr (r, −π/2) in the range −π < θ < 0 (see Figure 12.2) for the cases (λ1 , λ2 ) = (1, 2), (1, 0.5), and the isotropic case (1, 1), and comment on the results. 36.4. Show that the antiplane stress field near a traction-free notch −β < θ < β in a monoclinic material is characterized by a singular field with a dominant term proportional to r λ−1 , where π λ= 2γ

and

tan γ =

" 2 C44 C55 − C45 tan β (C44 + C45 tan β)

.

36.5. Use the Lekhnitskii formalism to find the stress components in Cartesian coördinates associated with the generalized Flamant solution, where a force F = {Fx , Fy , Fz }T per unit length acts along the line x = y = 0 on the otherwise tractionfree half-plane y > 0. Assume that the parameters pα , λα are known for the material of the half-plane. 36.6. Suppose that for a particular anisotropic material, equation (36.22) has a pair of repeated complex roots, so that p1 = p2 ≡ p = p3 . Show that a general solution

Problems

593

of the two dimensional elasticity problem can then be written in the form $ # φ = 2  f 1 (x + py) + (x + py) f 2 (x + py) + f 3 (x + p3 y) , Discuss how you would then find the corresponding function ψ, but do not complete the solution. 36.7. The half-plane y > 0 is loaded by a normal traction σ yy (x, 0) = S cos(λx) . Find expressions for the functions f α (ζα ) in equation (36.31) and hence show that the normal surface displacement is u y (x, 0) =

S D22 cos(λx) λ

where

D = ı AB −1 .

36.8. If the material is monoclinic — i.e. the stiffness matrix satisfies the condition (36.4) — show that the matrix equation (36.27) partitions into an in-plane problem for a(1) , a(2) and an antiplane problem for a(3) . Solve the resulting scalar equation for the antiplane eigenvalues p3 and show that they form a complex conjugate pair for all physically reasonable materials. 36.9. Use Maxwell’s reciprocal theorem (§38.1) and equation (36.39) to show that the matrix D is Hermitian. 36.10. Use an argument similar to that in §13.3 and the solution of §36.5.5 to find the energy-release rate for a plane crack as a function of the vector stress-intensity factor K = {K I , K I I , K I I I }T . 36.11. For a generally anisotropic material with non-uniform temperature T , the Fourier heat conduction law (14.10) and the constitutive law (1.42) take the more general form ∂T ∂u k qi = −K i j ; σi j = ci jkl + βi j T , ∂x j ∂xl where K is the conductivity matrix and β is the stress-temperature matrix, both of which are real and symmetric. If the temperature is steady-state and there are no internal heat sources, show that a general two dimensional temperature distribution can be written in the form T (x1 , x2 ) = 2 { f (ζT )}

where

ζT = x1 + κx2 ,

f is a holomorphic function and κ is a complex parameter. Then show that a particular solution for the thermoelastic displacement u can be written u = 2 {cg(ζT )} and find the governing equations for c and g(ζT ).

594

36 Anisotropic Elasticity

36.12. Show that the relations (36.32) imply that a, b satisfy the eigenvalue equations (36.36). 36.13. Find the strain components for a generally anisotropic material corresponding to the elementary state of stress where σzz = By is the only non-zero stress component. Then integrate the strain-displacement relations to find the corresponding displacements. In particular, by examining the terms that vary with z, show that this state of stress causes twist about the z-axis as well as bending about the x-axis, but that there is no bending about the y axis. Find the ratio between the twist per unit length and the curvature. 36.14. A solid circular cylinder of radius a with general anisotropy is loaded by a torque T . Find the twist per unit length β and the force and moment resultants on the ends needed to ensure that the axis of the cylinder remains straight. 36.15. Use Maxwell’s theorem (§38.1) to prove that the function h(θ) in equation (36.47) satisfies the condition h(θ + π) = h(θ). Then prove equation (36.48) by applying Betti’s reciprocal theorem (§38.2) to the concentrated force solution, using the two-dimensional sinusoidal solution of Problem 36.6 as auxiliary solution. 36.16. For the transversely isotropic material of equation (36.49), develop a potential function solution analogous to Solution E of Table 22.1. Start by assuming that the displacements can be represented in the form ux = 2

∂ψ ∂ψ ; u y = −2 ; uz = 0 , ∂y ∂x

determine the corresponding stress components, and hence find the equation that must be satisfied by the potential function ψ. Show that with a suitable scaling of the z-coördinate, this equation can be transformed into Laplace’s equation. 36.17. A transversely isotropic half-space z > 0 is loaded by a normal concentrated compressive force F applied at the origin. Find the surface displacements u r (r, 0), u z (r, 0), where r is distance from the origin. 36.18. A hexagonal close-packed crystal transforms into an identical pattern under a rotation of π/3 about an axis perpendicular to a basal plane. Show that this implies that the material is transversely isotropic in the basal plane.

Chapter 37

Variational Methods

Energy or variational methods have an important place in Solid Mechanics both as an alternative to the more direct method of solving the governing partial differential equations and as a means of developing convergent approximations to analytically intractable problems. They are particularly useful in situations where only a restricted set of results is required — for example, if we wish to determine the resultant force on a cross section or the displacement of a particular point, but are not interested in the full stress and displacement fields. Indeed, such results can often be obtained in closed form for problems in which a solution for the complete fields would be intractable. From an engineering perspective, it is natural to think of these methods as a consequence of the principle of conservation of energy or the first law of thermodynamics. However, conservation of energy is in some sense guaranteed by the use of Hooke’s law and the equilibrium equations. Once these physical premises are accepted, the energy theorems we shall present here are purely mathematical consequences. Indeed, the finite element method, which is one of the more important developments of this kind, can be developed simply by applying arguments from approximation theory (such as a least-squares fit) to the governing equations introduced in previous chapters. For this reason, these techniques are now more often referred to as Variational Methods, meaning that instead of seeking to solve the governing partial differential equations directly, we seek to define a scalar function of the physical parameters which is stationary (generally maximum or minimum) in respect to infinitesimal variations about the solution. In this chapter, we shall introduce only the principal theorems of this kind and indicate some of their applications1 . We shall present most of the derivations in the context of general anisotropy, using the index notation of §1.1.2, since this leads to a more compact presentation as well as demonstrating the generality of the results.

1

For a more comprehensive review of variational methods in elasticity, the reader is referred to S. G. Mikhlin, Variational Methods in Mathematical Physics, Pergamon, New York, 1964 and S. P. Timoshenko and J. N. Goodier, loc. cit., Chapter 8.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 J. R. Barber, Elasticity, Solid Mechanics and Its Applications 172, https://doi.org/10.1007/978-3-031-15214-6_37

595

596

37 Variational Methods

37.1 Strain Energy When a body is deformed, the external forces move through a distance associated with the deformation and hence do work. The fundamental premise of linear elasticity is that the deformation is linearly proportional to the load, so if the load is now removed, the deformation must pass through the same sequence of states as it did during the loading phase. It follows that the work done during loading is exactly recovered during unloading and must therefore be stored in the deformed body as strain energy. Suppose that the body Ω with boundary Γ is subjected to a body force distribution pi and boundary tractions ti = σi j n j (1.15), where n j defines a unit vector in the direction of the local outward normal to Γ . If the loads are applied sufficiently slowly for inertia effects to be negligible2 , and if the final elastic displacements are u i , the stored strain energy U in the body in the loaded state will be equal to the work W done by the external forces during loading, which is U=W =

1 2

 Ω

pi u i dΩ +

1 2

 Γ

ti u i dΓ .

(37.1)

37.1.1 Strain energy density The same principle can be applied to determine the strain energy in an infinitesimal rectangular element of material subjected to a uniform state of stress σi j . We conclude that the strain-energy density — i.e. the strain energy stored per unit volume — is given by 1 (37.2) U0 = σi j ei j . 2 Substituting for the stress or strain components from the generalized Hooke’s law (1.40, 1.42), we obtain the alternative expressions U0 =

∂u i ∂u k 1 1 1 ci jkl ei j ekl = ci jkl = si jkl σi j σkl . 2 2 ∂x j ∂xl 2

(37.3)

Notice incidentally that U0 must be positive for all possible states of stress or deformation and this places some inequality restrictions on the elasticity tensors ci jkl , si jkl . If the material is isotropic, these expressions reduce to

2

This implies that the stress field is always quasi-static in the sense of Chapter 7 and hence that all the work done by the external loads appears as elastic strain energy in the body, rather than partially as kinetic energy.

37.2 Conservation of Energy

597

 1  2 2 σx x + σ 2yy + σzz − 2ν(σ yy σzz + σzz σx x + σx x σ yy ) 2E   2 + 2(1 + ν) σ 2yz + σzx + σx2y   νe2 2 2 + ex2 x + e2yy + ezz + 2e2yz + 2ezx + 2ex2 y . =μ (1 − 2ν)

U0 =

(37.4)

37.2 Conservation of Energy The strain energy U stored in the entire body Ω can be obtained by summing that stored in each of its individual particles, giving  U=

Ω

U0 dΩ .

(37.5)

This expression and equation (37.1) must clearly yield the same result and hence 1 2

 Ω

1 pi u i dΩ + 2



 Γ

ti u i dΓ =

Ω

U0 dΩ .

(37.6)

This argument appeals to the principle of conservation of energy, but this principle is implicit in Hooke’s law, which guarantees that the loading is reversible. Thus, (37.6) can be derived from the governing equations of elasticity without explicitly invoking conservation of energy. To demonstrate this, we first substitute (1.15) into the second term on the left-hand side of (37.6) and apply the divergence theorem, obtaining 1 2

 Γ

  ∂ 1 1 n j σ ji u i dΓ = (σ ji u i )dΩ 2 Γ 2 Ω ∂x j   ∂σ ji ∂u i 1 1 u i dΩ + σ ji dΩ . = 2 2 ∂x j Ω ∂x j Ω

ti u i dΓ =

(37.7)

Finally, we use the equilibrium equation (2.4) in the first term on the right-hand side of (37.7) and Hooke’s law (1.42) in the second term to obtain 1 2

 Γ

ti u i dΓ = −

1 2

 Ω

pi u i dΩ +

1 2

 Ω

ci jkl

from which (37.6) follows after using (37.2) in the last term.

∂u k ∂u i dΩ , ∂xl ∂x j

598

37 Variational Methods

37.3 Potential Energy of the External Forces We can also construct a potential energy of the external forces which we denote by V . The reader is no doubt familiar with the concept of potential energy as applied to gravitational forces, but the definition can be extended to more general force systems as the work that the forces would do if they were allowed to return to their reference positions. For a single concentrated force F displaced through u this is defined as V = −F · u = −Fi u i . It follows by superposition that the potential energy of the boundary tractions and body forces is given by 

 V =−

Γt

ti u i dΓ −

Ω

pi u i dΩ ,

where Γt is that part of the boundary over which the tractions are prescribed. We can then define the total potential energy Π as the sum of the stored strain energy and the potential energy of the external forces — i.e. 1 Π ≡U +V = 2



∂u i ∂u k ci jkl dΩ − ∂x j ∂xl Ω



 Γt

ti u i dΓ −

Ω

pi u i dΩ . (37.8)

37.4 Theorem of Minimum Total Potential Energy Suppose that the displacement field u i satisfies the equilibrium equations (2.14) for a given set of boundary conditions and that we then perturb this state by a small variation δu i . The corresponding perturbation in Π is  δΠ =

Ω

ci jkl

∂δu i ∂u k dΩ − ∂x j ∂xl

 Γ

 ti δu i dΓ −

Ω

pi δu i dΩ ,

(37.9)

from (37.8). Notice that δu i = 0 in any region Γu of Γ in which the displacement is prescribed and hence the domain of integration Γt in the second term on the right-hand side of (37.8) can be replaced by Γ = Γu +Γt . Substituting for ti from (1.15) into the second term on the right-hand side of (37.9) and then applying the divergence theorem, we have

37.5 Approximate Solutions — the Rayleigh-Ritz Method

 Γ



599



 ∂  σi j δu i dΩ Γ Ω ∂x j   ∂σi j ∂δu i δu i dΩ + σi j dΩ . = Ω ∂x j Ω ∂x j

ti δu i dΓ =

σi j n j δu i dΓ =

Finally, using the equilibrium equation (2.4) in the first term and Hooke’s law (1.42) in the second, we obtain    ∂δu i ∂u k ti δu i dΓ = − pi δu i dΩ + ci jkl dΩ , ∂x j ∂xl Γ Ω Ω and comparing this with (37.9), we see that δΠ = 0. In other words, the equilibrium equation requires that the total potential energy must be stationary with regard to any kinematically admissible3 small variation δu i in the displacement field u i , or ∂Π =0. ∂u i

(37.10)

A more detailed second order analysis shows that the total potential energy must in fact be a minimum and this is intuitively reasonable, since if some variation δu i from the equilibrium state could be found which reduced Π , the surplus energy would take the form of kinetic energy and the system would therefore generally accelerate away from the equilibrium state. However, we emphasise that the above derivation of equation (37.10) makes no appeal to the principle of conservation of energy.

37.5 Approximate Solutions — the Rayleigh-Ritz Method We remarked in Chapter 2 that the problem of elasticity is to determine a stress field satisfying appropriate boundary conditions such that (i) the stresses satisfy the equilibrium equations (2.4) and (ii) the corresponding elastic strains satisfy the compatibility conditions (2.7). The compatibility conditions can be satisfied by defining the problem in terms of the displacements u i and we have just shown that the equilibrium equations are equivalent to the principal of minimum total potential energy. Thus, an alternative formulation is to seek a kinematically admissible displacement field u i such that the total potential energy Π is a minimum. This approach is particularly useful as a method of generating approximate solutions to problems that might otherwise be analytically intractable. Suppose that we approximate the displacement field in the body by the finite series u i (x1 , x2 , x3 ) =

n

Ak f i(k) (x1 , x2 , x3 ) ,

(37.11)

k=1

3

i.e. any perturbation that is a continuous function of position within the body and is consistent with the displacement boundary conditions.

600

37 Variational Methods

where the f i(k) are a set of approximating functions and Ak are arbitrary constants constituting the degrees of freedom in the approximation. The functions f i(k) must be continuous and single-valued and must also be chosen so as to ensure that equation (37.11) satisfies any displacement boundary conditions of the problem for all Ak , but they are otherwise unrestricted as to form. We can then substitute (37.11) into (37.8) and perform the integrals over Ω, obtaining the total potential energy as a quadratic function of the Ak . The theorem (37.10) demands that ∂Π = 0 ; k ∈ [1, n] , (37.12) ∂ Ak which defines n linear equations for the n unknown degrees of freedom Ak . The corresponding stress components can then be found by substituting (37.11) into Hooke’s law (1.42). If the approximating functions f i(k) are defined over the entire body Ω, this typically leads to series solutions (e.g. power series or Fourier series) and the method is known as the Rayleigh-Ritz method. It is particularly useful in structural mechanics applications such as beam problems, where the displacement is a function of a single spatial variable, but it can also be applied (e.g.) to the problem of a rectangular plate, using double Fourier series or power series.

Example To illustrate the method, we consider the problem of a rectangular plate −a < x < a, −b < y < b which makes frictionless contact with fixed rigid planes at y = ±b and which is loaded by compressive tractions σx x (±a, y) = −

Sy 2 b2

at x = ±a. The problem is symmetric about both axes, so u x must be odd in x and even in y, whilst u y is even in x and odd in y. Also, the displacement boundary condition requires that u y = 0 at y = ±b. The most general third order polynomial approximation satisfying these conditions is u x = A1 x 3 + A2 x y 2 + A3 x ; u y = A4 (b2 − y 2 ) and the corresponding strains are ex x =

∂u x = 3A1 x 2 + A2 y 2 + A3 ; ∂x ex y =

1 2



∂u y ∂u x + ∂x ∂y

e yy =

∂u y = −2 A4 y ∂y

 = A2 x y .

37.5 Approximate Solutions — the Rayleigh-Ritz Method

601

Substituting these results into (37.4) and then evaluating the integral (37.5), we obtain  U =



4μab(1 − ν)  (27A21 a 4 + 3A22 b4 + 15A23 + 20 A24 b2 15(1 − 2ν) −b −a  8μA2 a 3 b3 2 + 10 A2 A3 b2 + 30 A1 A3 a 2 + 10 A1 A2 a 2 b2 ) + . 9 b

a

U0 (x, y)d xd y =

The potential energy of the tractions on the boundaries x = ±a is  V =2

b

−b

Sy 2 (A1 a 3 + A2 ay 2 + A3 a) dy = 4Sab b2



A2 b2 A3 A1 a 2 + + 3 5 3

 ,

where the factor of 2 results from there being identical expressions for each boundary. Calculating Π =U + V and imposing the condition (37.12) for each degree of freedom Ai , i = 1, 4, we obtain four linear equations whose solution is A1 = 0 ; A3 = −

A2 = −

S(1 − 2ν) μ[5(1 − 2ν)a 2 + 2(1 − ν)b2 ]

5Sa 2 (1 − 2ν)2 ; 6μ(1 − ν)[5(1 − 2ν)a 2 + 2(1 − ν)b2 ]

A4 = 0 .

Rayleigh-Ritz solutions are conceptually straightforward, but they tend to generate lengthy algebraic expressions, as illustrated in this simple example. However, this is no bar to their use provided the algebra is performed in Mathematica or Maple. Also, if Fourier series are used in place of power series, the orthogonality of the corresponding integrals will often lead to significant simplifications. However, if high accuracy is required it is often more effective to use a set of piecewise continuous functions for the f i(k) (x1 , x2 , x3 ) of equation (37.11). The body is thereby divided into a set of elements and the displacement in each element is described by one or more shape functions multiplied by degrees of freedom Ak . Typically, the shape functions are defined such that the Ak represent the displacements at specified points or nodes within the body and the f i(k) are zero except in those elements contiguous to node k. They must also satisfy the condition that the displacement be continuous between one element and the next for all Ak . Once the approximation is defined, equation (37.12) once again provides n linear equations

602

37 Variational Methods

for the n nodal displacements. This is the basis of the finite element method 4 . Since the theorem of minimum total potential energy is itself derivable from Hooke’s law and the equilibrium equation, an alternative derivation of the finite element method can be obtained by applying approximation theory directly to these equations. To develop a set of n linear equations for the Ak , we substitute the approximate form (37.11) into the equilibrium equations, multiply by n weight functions, integrate over the domain Ω and set the resulting n linear functions of the Ak to zero. These equations will be identical to (37.12) if the weight functions are chosen to be identical to the shape functions.

37.6 Castigliano’s Second Theorem The strain energy U can be written as a function of the stress components, using the final expression in (37.3). We obtain  1 U= si jkl σi j σkl dΩ . 2 Ω We next consider the effect of perturbing the stress field by a small variation δσi j , chosen so that the perturbed field σi j + δσi j still satisfies the equilibrium equation (2.4) — i.e. ∂σi j + pi = 0 ∂x j

and

∂ (σi j + δσi j ) + pi = 0 . ∂x j

(37.13)

The corresponding perturbation in U will be  δU =

Ω

 si jkl σkl δσi j dΩ =

Ω

∂u i δσi j dΩ . ∂x j

(37.14)

The divergence theorem gives 

 Γ

 ∂  u i δσi j dΩ Ω ∂x j   ∂δσi j ∂u i δσi j dΩ + u i dΩ = Ω ∂x j Ω ∂x j

u i δσi j n j dΓ =

(37.15)

and by taking the difference between the two equations (37.13), we see that the second term on the right-hand side must be zero. Using (37.14, 37.15) and δσi j n j = δti , we then have 4

For more detailed discussion of the finite element method, see O. C. Zienkiewicz, The Finite Element Method, McGraw-Hill, New York, 1977, K. J. Bathe, Finite Element Procedures in Engineering Analysis, Prentice-Hall, Englewood Cliffs, NJ, 1982, T. J. R. Hughes, The Finite Element Method, Prentice Hall, Englewood Cliffs, NJ, 1987.

37.7 Approximations using Castigliano’s Second Theorem

 δU =

603



Γu

u i δσi j n j dΓ =

Γu

u i δti dΓ ,

where the integral is taken only over that part of the boundary Γu in which displacement u i is prescribed, since no perturbation in traction is permitted in Γt where ti is prescribed. It follows that the complementary energy  C ≡U−

Γu

u i ti dΓ

must be stationary with respect to all self-equilibrated variations of stress δσi j , or ∂C =0. ∂σi j

(37.16)

This is Castigliano’s second theorem. It enables us to state a second alternative formulation of the elasticity problem as the search for a state of stress satisfying the equilibrium equation (2.4), such that the complementary energy C is stationary. It is interesting to note that the theorem of minimum potential energy and Castigliano’s second theorem each substitutes for one of the two conditions equilibrium and compatibility identified in Chapter 2.

37.7 Approximations using Castigliano’s Second Theorem We recall that both the Airy stress function of Chapters 5–13 and the Prandtl stress function of Chapters 15, 17, 18 satisfy the equilibrium equations identically. In fact, these representations define the most general states of in-plane and antiplane stress respectively that satisfy the equilibrium equations without body force. Thus, if we describe the solution in terms of such a stress function, Castigliano’s second theorem reduces the problem to the search for a function satisfying the boundary conditions and equation (37.16). This leads to a very effective approximate method for both in-plane and antiplane problems in which the stress function φ is represented in the form n Ak f k (x, y) (37.17) φ(x, y) = k=1

by analogy with (37.11), where each of the functions f k satisfies the traction boundary conditions. This can generally be fairly easily achieved by making use of the mathematical description of the boundary line Γ . We shall start the discussion by considering the torsion problem of Chapter 17, for which this procedure is particularly straightforward. We shall then adapt it for the in-plane problem in §37.7.2.

604

37 Variational Methods

37.7.1 The torsion problem If the torsion problem is described in terms of Prandtl’s stress function ϕ, the tractionfree boundary condition for a simply-connected region Ω can be satisfied by requiring that ϕ = 0 on the boundary Γ . Suppose this boundary is defined by a set of m line segments f i (x, y) = 0, i ∈ [1, m]. Then the stress function ϕ = f 1 (x, y) f 2 (x, y)... f m (x, y)g(x, y)

(37.18)

will go to zero at all points on Γ for any function g(x, y). We can therefore choose a function g(x, y) with an appropriate number of arbitrary constants Ak , k ∈ [1, n] and then use Castigliano’s second theorem to develop n equations for the n unknowns. For example, the equations x = 0 and x 2 + y 2 −a 2 = 0 define the boundaries of a semicircular bar of radius a whose centre is at the origin. The stress function ϕ = x(x 2 + y 2 − a 2 )(A1 + A2 x) therefore satisfies the traction-free condition at all points on the boundary and contains two degrees of freedom, A1 , A2 . Once a function with an appropriate number of degrees of freedom has been chosen, we calculate the stress components σzx =

∂ϕ ∂ϕ ; σzy = − , ∂y ∂x

from (17.4) and hence the strain-energy density from (37.4), which in this case reduces to 

2 2 2 σzx + σzy 1 ∂ϕ 2 ∂ϕ U0 = = + . 2μ 2μ ∂x ∂y The torque corresponding to this stress distribution is given by equation (17.10) as  T =2

Ω

ϕdΩ

(37.19)

and hence the complementary energy per unit length of bar is  C=

U0 dΩ − T β

  2 2 ∂ϕ 1 ∂ϕ = + − 4μβϕ dΩ , 2μ Ω ∂x ∂y Ω

(37.20)

where β is the twist per unit length. Castigliano’s second theorem (37.16) then demands that

37.7 Approximations using Castigliano’s Second Theorem

605

∂C = 0 ; k ∈ [1, n] , ∂ Ak which defines n linear equations for the unknown constants Ak .

Example To illustrate the method, we consider a bar of square cross section defined by −a < x < a, −a < y < a. The boundaries are defined by the four lines x ± a = 0, y ± a = 0 and the simplest function of the form (37.18) is ϕ = A(x 2 − a 2 )(y 2 − a 2 ) , which has just one degree of freedom — the multiplying constant A. Substituting into equation (37.20) and evaluating the integrals, we have 1 C= 2μ



256A2 a 8 64μβ Aa 6 − 45 9

 .

To determine A, we impose the condition ∂C =0 ∂A

giving

A=

5μβ , 8a 2

after which the torque is recovered from (37.19) as T =

20μβa 4 . 9

The exact result can be obtained by substituting b = a in equation (17.25) and summing the series. The present one-term approximation underestimates the exact value by only about 1.3%. Closer approximations can be obtained by adding a few extra degrees of freedom, preserving the symmetry of the system with respect to x and y. For example, Timoshenko and Goodier5 give a solution of the same problem using the two degree of freedom function ϕ = (x 2 − a 2 )(y 2 − a 2 )[A1 + A2 (x 2 + y 2 )] , which gives a result for the torque within 0.15% of the exact value.

5

loc. cit. Art.111.

606

37 Variational Methods

37.7.2 The in-plane problem We showed in Chapter 4 that the most general in-plane stress field satisfying the equations of equilibrium (2.4) in the absence of body force can be expressed in terms of the scalar Airy stress function φ through the relations σx x =

∂2φ ∂2φ ∂2φ ; σ ; σ = − = . x y yy ∂ y2 ∂x∂ y ∂x 2

(37.21)

We also showed in §4.3.3 that the traction boundary-value problem — i.e. the problem of a two-dimensional body with prescribed tractions on the boundaries — has a solution in which the stress components are independent of Poisson’s ratio. We can therefore simplify the following treatment by considering the special case of plane stress (σzx = σzy = σzz = 0) with ν = 0 for which equation (37.4) reduces to U0 =

 2  σx x + σ 2yy

+

σx2y

4μ 2μ

2 2

2 2 2 2 ∂ φ ∂ φ ∂ φ 1 + +2 . = 4μ ∂x 2 ∂ y2 ∂x∂ y

We express the trial stress function φ in the form φ = φP + φH , where φ P is any particular function of x, y that satisfies the traction boundary conditions and φ H is a function satisfying homogeneous (i.e. traction-free) boundary conditions and that contains one or more degrees of freedom Ak . The stress components (37.21) will define a traction-free boundary if φ=0

and

∂φ =0, ∂n

(37.22)

on the boundary, where n is the local normal. In order to satisfy these conditions, we note that if f i (x, y) = 0 defines a line segment that is part of the boundary Γ , and if f i is a continuous function in the vicinity of the line segment, then we can perform a Taylor expansion about any point on the boundary. It then follows that the function [ f i (x, y)]2 will satisfy both conditions (37.22), since the first non-zero term in its Taylor expansion at the boundary will be quadratic. Thus, the conditions on φ H can be satisfied by defining  2 φ H = f 1 (x, y) f 2 (x, y)... f m (x, y) g(x, y)

(37.23)

37.7 Approximations using Castigliano’s Second Theorem

607

Fig. 37.1 The semi-circular plate.

where f i (x, y) = 0 i ∈ [1, m] define line segments comprising the boundary Γ and g(x, y) is any function containing one or more degrees of freedom.

Example Consider the semicircular plate of Figure 37.1 subjected to the self-equilibrated tractions

 3y 2 σx x = S 1 − 2 a on the straight boundary x = 0, the curved boundary r 2 − a 2 = x 2 + y 2 − a 2 = 0 being traction free. A particular solution maintaining the traction-free condition on the circle can be written φ P = B(r 2 − a 2 )2 = B(x 4 + y 4 + 2x 2 y 2 − 2a 2 x 2 − 2a 2 y 2 + a 4 ) , where B is a constant. The stress components σx x =

∂2φP ∂2φP 2 2 2 = −8Bx y = B(12y + 4x − 4a ) ; σ = − x y ∂ y2 ∂x∂ y

and on the boundary x = 0, σx x = B(3y 2 − a 2 ) ; σx y = 0 . Thus, the boundary conditions can be satisfied by the choice B=−

S S ; φ P = − 2 (r 2 − a 2 )2 . 2 4a 4a

608

37 Variational Methods

Notice that if the tractions on x = 0 had been higher order polynomials in y, we could have found an appropriate particular solution satisfying the traction-free condition on the curved edge using the form φ P = (r 2 − a 2 )2 g(x, y), where g(x, y) is a polynomial in x, y of appropriate order. The stress function φ P provides already a first order approximation to the stress field, but to improve it we can add the homogeneous term φ H = A1 x 2 (r 2 − a 2 )2 =

A1r 2 (r 2 − a 2 )2 (1 + cos 2θ) , 2

which satisfies traction-free conditions on both straight and curved boundaries in view of (37.23). The stress components in polar coördinates are σrr σθθ



r2 = S 1 − 2 + A1 (3r 4 − 4a 2 r 2 + a 4 ) + A1 (r 4 − a 4 ) cos 2θ a 

3r 2 = S 1 − 2 + A1 (15r 4 − 12a 2 r 2 + a 4 )(1 + cos 2θ) (37.24) a

σr θ = A1 (5r 4 − 6a 2 r 2 + a 4 ) sin 2θ from (8.8, 8.9) and the total strain energy is then calculated as U =

1 4μ



π/2



−π/2

a 0

 2  2 σrr + σθθ + 2σr2θ r dr dθ

 πa 2  14a 8 A21 − 5a 4 A1 S + 5S 2 = 30μ (using Mathematica or Maple). There are no displacement boundary conditions in this problem, so C =U and the degree of freedom A1 is determined from the condition ∂C =0 ∂ A1

or

A1 =

5S , 28a 4

after which the stress field is recovered by substitution in (37.24).

37.8 Uniqueness and Existence of Solution A typical elasticity problem can be expressed as the search for a displacement field u satisfying the equilibrium equations (2.14), such that the stress components defined through equations (1.42) and the displacement components satisfy appropriate boundary conditions. Throughout this book, we have tacitly assumed that this is a well-posed problem — i.e. that if the problem is physically well defined in the sense

37.8 Uniqueness and Existence of Solution

609

that we could conceive of loading a body of the given geometry in the laboratory, then there exists one and only one solution. To examine this question from a mathematical point of view, suppose provisionally that the solution is non-unique, so that there exist two distinct stress fields σ 1 , σ 2 , both satisfying the field equations and the same boundary conditions. We can then construct a new stress field Δσ = σ 1 − σ 2 by taking the difference between these fields, which is a form of linear superposition. The new field Δσ clearly involves no external loading, since the same external loads were included in each of the constituent solutions ex hyp. We therefore conclude from (37.1) that the corresponding total strain energy U associated with the field Δσ must be zero. However, U can also be written as a volume integral of the strain-energy density U0 as in (37.5), and U0 must be everywhere positive or zero. The only way these two results can be reconciled is if U0 is zero everywhere, implying that the stress is everywhere zero, from (37.3). Thus, the difference field Δσ is null, the two solutions must be identical contra hyp. and only one solution can exist to a given elasticity problem. The question of existence of solution is much more challenging and will not be pursued here. A short list of early but seminal contributions to the subject is given by Sokolnikoff 6 who states that “the matter of existence of solutions has been satisfactorily resolved for domains of great generality.” More recently, interest in more general continuum theories including non-linear elasticity has led to the development of new methodologies in the context of functional analysis7 .

37.8.1 Singularities In §11.2.1, we argued that stress singularities are acceptable in the mathematical solution of an elasticity problem if and only if the strain energy in a small region surrounding the singularity is bounded. In the two-dimensional case (line singularity), this leads to equation (11.17) and hence to the conclusion that the stresses can vary with r a as r → 0 only if a > −1. We can now see the reason for this restriction, since if there were any points in the body where the strain energy was not integrable, the total strain energy (37.5) would be ill-defined and the above proof of uniqueness would fail. If these restrictions on the permissible strength of singularities are not imposed, it is quite easy to generate examples in which a given set of boundary conditions permits multiple solutions. Consider the flat punch problem of §12.5.2 for which the contact pressure distribution is given by equation (12.30). If we differentiate the stress and displacement fields with respect to x, we shall generate a field in which the contact traction is

6

I. S. Sokolnikoff, Mathematical Theory of Elasticity, McGraw-Hill, New York, 2nd.ed. 1956, §27. See for example, J. E. Marsden and T. J. R. Hughes, Mathematical Foundations of Elasticity, Prentice-Hall, Englewood Cliffs, 1983, Chapter 6.

7

610

37 Variational Methods

p(x) =

d dx



Fx F =− − √ ; −a < x < a π(a 2 − x 2 )3/2 π a2 − x 2

and the normal surface displacement is u y (x, 0) =

d (C) = 0 ; −a < x < a dx

from (12.29). The traction-free condition is retained in |x| > a, y = 0. We can now add this differentiated field to the solution of any frictionless contact problem over the contact region −a < x < a without affecting the satisfaction of the displacement boundary condition (12.17), so solutions of such problems become non-unique. However, the superposed field clearly involves stresses varying with r 3/2 near the singular points (±a, 0) and if (non-integrable) singularities of this strength are precluded, uniqueness is restored. The three-dimensional equivalent of equation (11.17) would involve a stress field that tends to infinity with R a as we approach the point R = 0. In this case, the integral in spherical polar coördinates centred on the singular point would take the form U=

1 2

2π π R

 0

0

0



R

σi j ei j R 2 sin βd Rdβdθ = C

R 2a+2 d R ,

0

where C is a constant. This integral is bounded if and only if a > − 23 . In both two and three-dimensional cases, if a concentrated force F is applied to the body, either at a point on the boundary or at an interior point, the resulting stress field violates this energy criterion. An ‘engineering’ argument can be made that the real loading in such cases consists of a distribution of pressure or body force over a small finite region A that is statically equivalent to F, in which case no singularity is involved. Furthermore, a unique solution is obtained for each member of a regular sequence of such problems in which A is progressively reduced. Thus, if we conceive of the point force as the limit of this set of distributions, the uniqueness theorem still applies. Saint-Venant’s principle §3.1.2 implies that only the stresses close to A will be changed as A is reduced, so the concentrated force solution also represents an approximation to the stress field distant from the loaded region when A is finite. Sternberg and Eubanks8 proposed an extension of the uniqueness theorem to cover these cases. Similar arguments can be applied to the stronger singularity due to a concentrated moment. However, there is now some ambiguity in the limit depending on the detailed sequence of finite states through which it is approached9 . 8

E. Sternberg and R. A. Eubanks (1955), On the concept of concentrated loads and an extension of the uniqueness theorem in the linear theory of elasticity, Journal of Rational Mechanics and Analysis, Vol. 4, pp. 135–168. 9 E. Sternberg and V. Koiter (1958), The wedge under a concentrated couple: A paradox in the two-dimensional theory of elasticity, ASME Journal of Applied Mechanics, Vol.25, pp. 575–581; X. Markenscoff (1994), Some remarks on the wedge paradox and Saint Venant’s principle, ASME Journal of Applied Mechanics, Vol.61, pp. 519–523.

Problems

611

Problems 37.1. For an isotropic material, find a way to express the strain-energy density U0 of equation (37.4) in terms of the stress invariants of equations (1.20–1.22). 37.2. An annular ring a